JP3905375B2 - Target tracking device - Google Patents

Description

【００６０】
の様に推定する。ここで、β_FT、Ｐ_dkは事前に設定するパラメータである。これに対して、クラスタの体積Ｖ_kは、クラスタに属する全ての航跡を統合し、統合予測値、統合予測誤差共分散を以下のやり方に従って計算する。
【００６１】
クラスタ中にＭ個の航跡があり、それらをＴ_i（ｉ＝１，・・・，Ｍ）、各々の航跡信頼度をβ_i（ｉ＝１，・・・，Ｍ）とする。また、航跡Ｔ_iの予測ベクトルをＸ_k,iアンダーバー（−）、予測誤差共分散行列をＰ_k,i（−）とする。このとき、Ｍ個の航跡を統合してできる航跡の予測値は（２）式、予測誤差共分散行列は（３）式で計算される。
【００６２】
【数２】

【数３】

【００６３】
これらの統合予測値、統合予測誤差共分散行列から統合ゲートを求め、その体積でクラスタの体積を近似する。
ここで統合ゲートとは、
【００６４】
【数４】

【００６５】
によって求めた残差共分散行列を使った条件式（５）を満たすＺアンダーバーがとりうる領域である。
【００６６】
【数５】

【００６７】
（２）仮説の航跡数毎の信頼度による推定
各仮説を航跡数毎に分類する。分類毎に仮説の信頼度の和を出し、その和が最も大きい分類における航跡数を、目標数とする。
【００６８】
以下、例を用いて説明する。以下の５仮説（「ＩＤ：｛航跡組｝信頼度」と表記）が生成されていたとする。
Ｈ１：｛Ｔ１｝ｒｅｌ１
Ｈ２：｛Ｔ１，Ｔ３，Ｔ６｝ｒｅｌ２
Ｈ３：｛Ｔ２，Ｔ５｝ｒｅｌ３
Ｈ４：｛Ｔ２，Ｔ３，Ｔ６｝ｒｅｌ４
Ｈ５：｛Ｔ５，Ｔ６｝ｒｅｌ５
【００６９】
仮説が参照する航跡数に着目して、
航跡数１のグループ：Ｈ１， 信頼度ｒｅｌ１
航跡数２のグループ：Ｈ３，Ｈ５， 信頼度ｒｅｌ３＋ｒｅｌ５
航跡数３のグループ：Ｈ２，Ｈ４， 信頼度ｒｅｌ２＋ｒｅｌ６
【００７０】
この３つグループのうち、航跡数２の仮説によるグループの信頼度が最も高かったとする。このとき、現サンプリング時刻のサンプリング別目標数の推定値は２となる。
【００７１】
（３）航跡数の信頼度による重み付け平均による推定
クラスタ中のｐ個の仮説をＨ_i（ｉ＝１，・・・，ｐ）、各々の信頼度をβ_i（ｉ＝１，・・・，ｐ）、さらに航跡数をｔ_i（ｉ＝１，・・・，ｐ）とする。このとき、現サンプリング時刻のサンプリング別目標数の推定値は（６）式によって求める。
【００７２】
【数６】

【００７３】
これは航跡数の仮説信頼度による加重平均である。
【００７４】
次に目標数推定時期判定部９２が、追尾状況が航跡数制御のための目標数決定条件を満たすか否か判定する。この判定のための条件は以下の２つの何れかとする。
（１）現サンプリング時刻のサンプリング別目標数の推定値と前サンプリング時刻におけるサンプリング別目標数の推定値の差が、あらかじめ設定した閾値より小さい場合。
（２）前回の航跡数制御用目標数決定時刻から経過したサンプル数が閾値に達した場合。
【００７５】
この目標数決定条件を満たすと判定された場合、航跡数制御用目標数決定部９３がステップＳ００６２「航跡数制御用目標数決定」処理、続いて仮説中航跡制限部９４がステップＳ００６３「仮説中航跡制限」を実行する。目標数決定条件が満たされない場合、何もせずに、次のステップＳ００７「準最適化とクラスタ統合」処理へと移行する。
【００７６】
ステップＳ００６２「航跡数制御用目標数決定」では、サンプリング別目標数推定値の時系列より、航跡数制御用目標数を算出する。この算出方法は、以下の２つの何れかとする。
【００７７】
（１）全体の平均による算出
この段階以前に実行された航跡数制御用目標数決定に使用されていない、サンプリング別目標数推定値をＮ_i＾（ｉ＝１，・・・ｒ）とする。この時、目標数は以下の（７）式で決定される。
【００７８】
【数７】

【００７９】
（２）フィルタの平滑処理による算出
追尾処理開始時刻からのサンプリング時刻毎のサンプリング別目標数推定値の履歴を入力とした、フィルタの平滑値により算出する。以下はαフィルタを用いた計算式である。
【００８０】
【数８】

【数９】

【００８１】
ここで、Ｎ_skはサンプリング時刻ｔ_kにおける目標数の平滑値、Ｎ_pkは目標数の予測値、Ｎ_k＾はフィルタに入力する、サンプリング時刻ｔ_kにおけるサンプリング別目標数推定値である。また、α_kはパラメータである。
【００８２】
ステップＳ００６３「仮説中航跡数制限」では、仮説中航跡制限部９４によって、現サンプリング時刻における仮説群の中から、その航跡数が前ステップで決定された航跡数制御用目標数に一致する仮説のみを残し、一致しない仮説を全て削除する。
【００８３】
次のステップＳ００７「準最適化とクラスタ分離」において、仮説縮小部５が信頼度の低い仮説の削除、似た仮説の統合によって仮説群を縮小する準最適化処理を実行し、さらにクラスタ分離部６がそれに伴うクラスタ分離の処理を実行する。
【００８４】
この準最適化処理では、次ステップＳ００８「本航跡判定」において本航跡の昇格が起こりにくい状況である場合、「過去Ｎ回の時刻分における探知データの相関内容が同一の仮説を統合する」仮説削減において、パラメータＮを時刻が進むにつれて小さく設定する様にしてもよい。この設定で仮説削減を行うことにより、「本航跡判定」において本航跡をより速く検出することが可能となる。
【００８５】
次のステップＳ００８「本航跡判定」において、航跡決定部８が仮航跡群の中から、目標を追跡している可能性が極めて高い仮航跡を本航跡に昇格させる処理が行われる。
【００８６】
このように本実施の形態においては、クラスタ内に存在する目標数を探知データの数、または仮説中の航跡数を基に推定し、仮説中の航跡数の制限に利用するので、目標数の特定が速まり、それにより目標航跡の検出を効率的に行うことが可能となる。また、誤った航跡数を持つ仮説を排除できるので、目標航跡の検出をより正確に行うことができる。
【００８７】
実施の形態２．
図５はこの発明の実施の形態２の目標追尾装置を示す全体構成図である。図６は図５の仮説生成用目標数推定部を詳細に示す図である。図７は図５に示された仮説生成用目標推定部の動作を示すフローチャートである。図８は図７の仮説生成用目標推定ステップを詳細に示すフローチャートである。
【００８８】
本実施の形態の目標追尾装置は、図５に示されるように、クラスタ中の目標数を推定し、生成する仮説中の航跡数を決定する仮説生成用目標数推定部１０を備えている。また、仮説生成用目標数推定部１０は、図６に示されるように、サンプリング時刻毎に目標数の推定を行うサンプリング別目標数推定部１０１と、準最適化に使用する目標数の決定を行うか否かを判断する目標数推定時期判断部と１０２、サンプリング時刻毎の目標数の履歴より航跡数制御用目標数を推定する航跡数制御用目標数決定部１０３と、航跡相関行列毎に、更新仮説が持つべき航跡数を決定する航跡相関行列制限部１０４とを有している。
その他の構成は概略実施の形態１と同様である。
【００８９】
次に、図７及び図８に沿って、本実施の形態の動作を説明する。
まず、ステップＳ０１１「次サンプリングデータの読込」において、探知データ選択部０がその時刻の探知データを、センサから読み込む。
【００９０】
次のステップＳ０１２「ゲート内判定」において、探知データ選択部０が既存航跡のゲートにおける探知データの内外判定を行って相関状況を検査し、入ってきた探知データの属するクラスタを決定する。
【００９１】
次のステップＳ０１３「航跡および新クラスタの作成」においては、クラスタ新設統合部１により、既存航跡ゲートと探知データの相関具合を元に航跡が作成される。作成される航跡は以下の更新航跡、新航跡、メモリトラック航跡の３つの種類がある。
【００９２】
また、本ステップＳ０１３では、どのクラスタとも相関がなかった探知データについては、その探知データを起点とする新たなクラスタを生成する。また、本ステップＳ０１３ではこの時刻における探知データにより同値関係が複数のクラスタ間で発生した場合、それらのクラスタを統合する処理が行われる。
【００９３】
次のステップＳ０１４「クラスタ毎の航跡相関行列生成」においては、まず、ゲート内判定行列算出部２によってゲート内判定行列が生成される。ゲート内判定行列は、探知データと航跡の可能な相関関係を示す表である。
【００９４】
次に、航跡相関行列算出部３がこのゲート内判定行列から航跡相関行列を作る。航跡相関行列は、
・１となる要素と同一のゲート内判定行列の要素は必ず１となっている。
・各行で、１となる要素の数は１つのみ。
・各列は高々１つの要素が１となる。ただし、第１列はいくつ１の要素があってもよい。
という条件を満たす全ての行列である。これは、各探知データの互いに矛盾しない解釈の組み合わせである。
ここまでの処理は、実施の形態１と概略同様である。
【００９５】
次のステップＳ０１５「仮説生成用目標数推定」において、仮説生成用目標数推定部１０が目標数推定処理を行う。この仮説生成用目標数推定部１０の動作を図６及び図８に沿って説明する。まず、ステップＳ０１５１「サンプリング別目標数計算」において、サンプリング別目標数推定部１０１が、最新サンプリング時刻における追尾に関する情報を基に、サンプリング別目標数の推定計算を行う。その方法を以下に説明する。なお、これは実施の形態１における同名のステップにおける「（１）入力された探知データ数を用いた推定方法」と全く同様の方法である。
【００９６】
サンプリング時刻ｔ_kにおけるあるクラスタに相関した探知データ数をｍ_kとする。クラスタの体積をＶ_k、誤警報密度をβ_FT、探知確率をＰ_dkとする。これらのパラメータより、現サンプリング時刻の推定目標数は、
【００９７】
【数１０】

【００９８】
の様に推定する。ここで、β_FT、Ｐ_dkは事前に設定するパラメータである。これに対して、クラスタの体積Ｖ_kは、クラスタに属する航跡群を統合し、統合予測値、統合予測誤差共分散を計算し、さらにこれらの統合予測値、統合予測誤差共分散行列から統合ゲートを求め、その体積でクラスタの体積を近似する。
【００９９】
次に、目標数推定時期判定部１０２が、現時刻の追尾状況が目標数決定条件を満たすか否か判定する。この判定のための条件は以下の２つの何れかとする。なお、これは実施の形態１における同名の判定ステップにおける（１）（２）と全く同等の方法である。
【０１００】
（１）現サンプリング時刻のサンプリング別目標数の推定値と前サンプリング時刻におけるサンプリング別目標数の推定値の差が、あらかじめ設定した閾値より小さい場合。
（２）前回の航跡数制御用目標数決定時刻から経過したサンプル数が閾値に達した場合。
【０１０１】
この目標数決定条件を満たすと判定された場合、航跡数制御用目標数決定部１０３が、ステップＳ０１５２「航跡数制御用目標数決定」処理、続いて、航跡相関行列制限部１０４が、ステップＳ０１５３「航跡相関行列制限」を実行する。目標数決定条件が満たされない場合、何もせずに、次のステップＳ０１６「仮説の生成」へと移行する。
【０１０２】
ステップＳ０１５２「航跡数制御用目標数決定」では、サンプリング時刻毎に計算された目標数推定値より、航跡数制御用目標数を算出する。この算出方法は、以下の２つの何れかとする。なお、これは実施の形態１における同名のステップにおける（１）（２）と全く同等のやり方である。
【０１０３】
（１）全体の平均による算出
この段階以前に実行された航跡数制御用目標数決定に使用されていない、サンプリング別目標数推定値をＮ_i＾（ｉ＝１，・・・）とする。この時、目標数は以下の（７）式で決定される。
【０１０４】
【数１１】

【０１０５】
（２）フィルタの平滑処理による算出
追尾処理開始時刻からのサンプリング時刻毎のサンプリング別目標数推定値の時系列を入力とした、フィルタの平滑値により算出する。
【０１０６】
ステップＳ０１５３「航跡相関行列制限」では、航跡相関行列制限部１０４によって、前ステップＳ０１４「クラスタ毎の航跡相関行列生成」において生成された全ての航跡相関行列について、必要な既存航跡必要数を登録する。この既存航跡必要数は、
【０１０７】
航跡数制御用目標数推定値−航跡相関行列中の新目標部の”１”が設定された要素数
【０１０８】
である。航跡相関行列がｎ行ｍ列で構成されていたとしたら、航跡相関行列の新目標部とはその第ｍ−ｎ＋１列〜第ｍ列の部分である。目標推定値が２と決定されたとする。図１４の航跡相関行列の例で、各々の既存航跡必要数は以下のとおりとなる。
【０１０９】
（１）→２
（２）→１
（３）→１
（４）→２
（５）→１
（６）→−１
【０１１０】
ここで算出された既存航跡必要数は、航跡相関行列と組み合わせて更新仮説を作ることが可能な既存仮説が持つべき航跡数である。また、この既存航跡必要数が０未満の値になったら、その航跡相関行列は削除する。図１４の例では（６）の航跡相関行列は削除される。
【０１１１】
次のステップＳ０１６「仮説の作成」において、仮説更新部４が、既存仮説を航跡相関行列によって更新する処理を実行する。各仮説の更新では、その中に含まれている航跡以外の航跡の存在を仮定している航跡相関行列を除いた全ての航跡相関行列を使用して、可能な仮説と航跡相関行列の組み合わせによって既存仮説を更新し、新たな仮説を生成する。
【０１１２】
現サンプリング時刻で、ステップＳ０１５２の処理で航跡数制御用目標数の推定が行われていた場合、本ステップで既存仮説の更新が可能となるのは、組み合わせる航跡相関行列の既存航跡必要数と既存仮説中の航跡数が一致する場合のみである。既存航跡必要数は、ステップＳ０１５３「航跡相関行列制限」で航跡相関行列毎に計算し、登録されている。図１４の航跡相関行列については、
【０１１３】
（１）からは既存航跡数２の既存仮説のみが組み合わせ可能
（２）からは既存航跡数１の既存仮説のみが組み合わせ可能
（３）からは既存航跡数１の既存仮説のみが組み合わせ可能
（４）からは既存航跡数２の既存仮説のみが組み合わせ可能
（５）からは既存航跡数１の既存仮説のみが組み合わせ可能
となる。
また、既存航跡の数が航跡数制御用目標数推定値よりも多い場合は、この既存仮説からは更新仮説は作らない。
【０１１４】
次のステップＳ０１７「準最適化とクラスタ分離」において、仮説縮小部５が信頼度の低い仮説の削除、似た仮説の統合によって仮説群を縮小する準最適化処理を実行し、さらに、クラスタ分離部６がそれに伴うクラスタ分離の処理を実行する。
【０１１５】
次のステップＳ０１８「本航跡判定」において、航跡決定部８が仮航跡群の中から、目標を追跡している可能性が極めて高い仮航跡を本航跡に昇格させる処理が行われる。
【０１１６】
このように本実施の形態においては、クラスタ内に存在する目標数を探知データの数を基に決定し、仮説生成時の航跡数の制限に利用するので、目標数の特定が速まり、それにより目標航跡の検出を効率的に行うことが可能となる。また、誤った航跡数を持つ仮説を排除できるので、目標航跡の検出をより正確に行うことができる。
【０１１７】
【発明の効果】
この発明に係る目標追尾装置は、各クラスタに含まれる既存航跡から探知データの存在可能領域を算出するゲート算出部と、前記存在可能領域と全体の探知データからクラスタ毎に航跡対応の探知データを選択する探知データ選択部と、既存のクラスタ内の航跡と対応がとれない探知データについて新クラスタを作成し、また、複数のクラスタ内の航跡と対応付けられる探知データがある場合に対応するクラスタを統合するクラスタ新設・統合部と、既存航跡と探知データの相関の有無よりゲート内判定行列を作成するゲート内判定行列算出部と、前記ゲート内判定行列に基づいて、各探知データに対する解釈の可能な組合せを全て抽出し航跡相関行列を算出する航跡相関行列算出部と、既存の仮説と前記航跡相関行列を組み合わせて新たに仮説を生成、更新する仮説更新部と、前記仮説から信頼度の低い仮説の削減によって仮説群を縮小する準最適化処理を実行する仮説縮小部と、航跡間の探知データの共有の有無に基づいてクラスタを分離するクラスタ分離部と、前記航跡から所定の条件に基づいて目標である可能性の高い航跡を選択して本航跡とする航跡決定部と、クラスタ中の目標数を推定し、前記準最適化処理の内容を決定する準最適化用目標数推定部とを備えた目標追尾装置であって、前記準最適化用目標数推定部は、サンプリング時刻毎に前記目標数の推定を行うサンプリング別目標数推定部と、前記準最適化処理に使用する前記目標数の決定を行うか否かを判断する目標数推定時期判断部と、サンプリング時刻毎の前記目標数の履歴より航跡数制御用目標数を推定する航跡数制御用目標数決定部と、前記航跡数制御用目標数決定部が推定した前記航跡数制御用目標数に基づいて仮説の削除を行う仮説中航跡数制限部とを有する。そのため、クラスタ内に存在する目標数を探知データの数、または仮説中の航跡数を基に推定し、仮説中の航跡数の制限に利用するので、目標数の特定が速まり、それにより目標航跡の検出を効率的に行うことが可能となる。また、誤った航跡数を持つ仮説を排除できるので、目標航跡の検出をより正確に行うことができる。
【０１１８】
また、この発明に係る目標追尾装置は、各クラスタに含まれる既存航跡から探知データの存在可能領域を算出するゲート算出部と、前記存在可能領域と全体の探知データからクラスタ毎に航跡対応の探知データを選択する探知データ選択部と、既存のクラスタ内の航跡と対応がとれない探知データについて新クラスタを作成し、また、複数のクラスタ内の航跡と対応付けられる探知データがある場合に対応するクラスタを統合するクラスタ新設・統合部と、既存航跡と探知データの相関の有無よりゲート内判定行列を作成するゲート内判定行列算出部と、前記ゲート内判定行列に基づいて、各探知データに対する解釈の可能な組合せを全て抽出し航跡相関行列を算出する航跡相関行列算出部と、既存の仮説と前記航跡相関行列を組み合わせて新たに仮説を生成、更新する仮説更新部と、前記仮説から信頼度の低い仮説の削減によって仮説群を縮小する準最適化処理を実行する仮説縮小部と、航跡間の探知データの共有の有無に基づいてクラスタを分離するクラスタ分離部と、前記航跡から所定の条件に基づいて目標である可能性の高い航跡を選択して本航跡とする航跡決定部と、クラスタ中の目標数を推定し、生成する仮説中の航跡数を決定する仮説生成用目標数推定部とを備えた目標追尾装置であって、前記仮説生成用目標数推定部は、サンプリング時刻毎に前記目標数の推定を行うサンプリング別目標数推定部と、前記準最適化処理に使用する前記目標数の決定を行うか否かを判断する目標数推定時期判断部と、サンプリング時刻毎の前記目標数の履歴より航跡数制御用目標数を推定する航跡数制御用目標数決定部と、前記航跡相関行列毎に、更新仮説が持つべき航跡数を決定する航跡相関行列制限部とを有する。そのため、クラスタ内に存在する目標数を探知データの数を基に決定し、仮説生成時の航跡数の制限に利用するので、目標数の特定が速まり、それにより目標航跡の検出を効率的に行うことが可能となる。また、誤った航跡数を持つ仮説を排除できるので、目標航跡の検出をより正確に行うことができる。
【０１１９】
また、サンプリング別目標数推定部は、サンプリング時刻毎の目標数を、クラスタ中の航跡より算出されるクラスタの体積とクラスタ内で相関した最新サンプリング時刻における探知データ数を用いて計算する。すなわち、クラスタ中の航跡より算出されるクラスタの体積とクラスタ内で相関した探知データの数を用いてサンプリング別目標数を計算するため、観測状況を反映した正確な目標数推定が可能となる。
【０１２０】
また、サンプリング別目標数推定部は、サンプリング時刻毎の目標数を、仮説を航跡数毎に分類してその信頼度の和が最も大きな値となる航跡数として推定する。すなわち、仮説を航跡数毎に分類してその信頼度の和が最も大きな値となる航跡数を、サンプリング別目標数として推定するため、追尾装置が維持する仮説の内容を反映した目標数推定が可能となる。
【０１２１】
また、サンプリング別目標数推定部は、サンプリング時刻毎の目標数を、仮説の航跡数を信頼度で重み付けしてその平均値として推定する。すなわち、仮説の航跡数を信頼度で重み付けしてその平均値をサンプリング別目標数として推定するため、追尾装置が維持する仮説の内容を反映した正確な目標数推定が可能となる。
【０１２２】
また、目標数推定時期判断部は、現サンプリング時刻の推定目標数と前サンプリング時刻における推定目標数の差を目標数推定時期の判断基準とする。すなわち、現サンプリング時刻の推定目標数と前サンプリング時刻における推定目標数の差を目標数決定時期の判断基準とするため、目標数推定が安定した時期に、目標数決定を行うことが可能となり、正確な目標数決定が行える。
【０１２３】
また、目標数推定時期判断部は、前回の目標数決定時刻から経過したサンプル数を目標数推定時期の判断基準とする。すなわち、前回の目標数決定時刻から経過したサンプル数を目標数決定時期の判断基準とするため、目標数決定が定期的に行われ、いつまで経っても目標数が決定できないという事態にならない。
【０１２４】
また、航跡数制御用目標数決定部は、サンプリング時刻毎の目標数推定値の平均値を航跡数制御用目標数と推定する。すなわち、サンプリング時刻毎の目標数推定値の平均値を目標数と推定するために、各サンプリング時刻毎の目標数を反映した正確な目標数推定が可能となる。
【０１２５】
また、航跡数制御用目標数決定部は、目標数推定値の履歴を入力としたフィルタの平滑値を航跡数制御用目標数と推定する。すなわち、目標数推定値の履歴を入力としたフィルタの平滑値として推定するため、各サンプリング時刻毎の目標数が時系列として考慮され、より正確な目標数推定が可能となる。
【０１２６】
また、仮説縮小部は、航跡決定部の処理で本航跡の昇格が所定の程度に起こりにくい状況である場合、「過去Ｎ回の時刻分における探知データの相関内容が同一の仮説を統合する」仮説削減方法において、パラメータＮを時刻が進むにつれて小さく設定する。すなわち、「過去Ｎ回の時刻分における探知データの相関内容が同一の仮説を統合する」仮説削減方式において、パラメータＮを時刻が進むにつれて小さく設定するために、目標航跡が航跡確立の判定基準を満たし易くなり、早期に目標航跡を検出することができる。
【図面の簡単な説明】
【図１】 この発明の実施の形態１の目標追尾装置を示す全体構成図である。
【図２】 図１の準最適化用目標数推定部を詳細に示す図である。
【図３】 図１に示された目標追尾装置の動作を示すフローチャートである。
【図４】 図３の準最適化用目標数推定ステップを詳細に示すフローチャートである。
【図５】 この発明の実施の形態２の目標追尾装置を示す全体構成図である。
【図６】 図５の仮説生成用目標数推定部を詳細に示す図である。
【図７】 図５に示された仮説生成用目標推定部の動作を示すフローチャートである。
【図８】 図７の仮説生成用目標推定ステップを詳細に示すフローチャートである。
【図９】 従来の目標追尾装置を示す全体構成図である。
【図１０】 図９に示された従来の目標追尾装置の動作を示すフローチャートである。
【図１１】 各時刻でサンプリングされた探知データの相関具合を説明する説明図である。
【図１２】 各時刻でサンプリングされた探知データの相関具合を説明する説明図である。
【図１３】 各時刻における航跡と探知データの相関の例を示す説明図である。
【図１４】 ゲート内判定行列とそれより得られる航跡相関行列（全体の一部６個）の例を示す図である。
【符号の説明】
０ 探知データ選択部、１ クラスタ新設・統合部、２ ゲート内判定行列算出部、３ 航跡相関行列算出部、４ 仮説更新部、５ 仮説縮小部、６ クラスタ分離部、７ ゲート算出部、８ 航跡決定部、９ 準最適化用目標数推定部、９１ サンプリング別目標数推定部、９２ 目標数推定時期判断部、９３ 航跡数制御用目標数決定部、９４ 仮説中航跡数制限部、１０ 仮説生成用目標数推定部、１０１ サンプリング別目標数推定部、１０２ 目標数推定時期判断部、１０３ 航跡数制御用目標数決定部、１０４ 航跡相関行列制限部。[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a target tracking device that estimates the trajectories of a plurality of targets from detection data obtained from a sensor such as a radar, and is particularly effective and efficient in a tracking initialization function for detecting unknown targets. The present invention relates to a target tracking device that can be processed automatically.
[0002]
[Prior art]
In target tracking for estimating a track as a target trajectory using detection data obtained from a sensor, it is first determined whether the detection data is from a target or an unnecessary signal. If it is determined that the target is from a target, it is determined whether it is a target that has already been tracked (if there are a plurality of existing targets, which one corresponds to it) or a newly generated target.
[0003]
For this determination, a gate that is a spatial region centered on a predicted value obtained by calculating a position at a corresponding time from the speed of an existing target is used. The size of the gate is set so that the probability that the detection data from the tracking target is observed in the gate will be a high value such as 0.9 or more. The detection data that has entered the gate is likely to be from an existing target, and the detection data that has not entered any gate is highly likely to be an unnecessary signal or a new target.
[0004]
For detection data determined to have been obtained from an existing track, the estimated true position and speed of the target at the time when the detection data was detected using the detection data information and the predicted value of the existing track. Calculate and update existing tracks.
[0005]
FIG. 13 shows an example of the correlation between the track and the detection data. At the current time t = 3, four pieces of detection data 30, 31, 32, 33 are obtained. Detection data 30 and 31 enter the gate of the track of the target 0 (track composed of the detection data 00, 10, and 20), and the gate of the track of the target 1 (the track composed of the detection data 01, 11, and 21) The detection data 32 is entered in. The detection data 33 does not enter either gate.
[0006]
As one possibility, there is an assignment of “detection data 30: target 0, detection data 31: unnecessary signal, detection data 32: target 1, detection data 33: new target”, and another possibility is “detection data”. 30: Unnecessary signal, detection data 31: target 0, detection data 32: target 1, detection data 33: unnecessary signal ”is also considered.
[0007]
In such a situation where there is a plurality of assignment possibilities, if any one is determined, there is a high risk of failure to track the target if it is an error. Therefore, more accurate target tracking is possible by maintaining a plurality of such assignments (hypotheses), determining correlation information between detection data and targets for each hypothesis, and finally leaving the best hypothesis.
[0008]
A target tracking technique based on this method is disclosed, for example, in “Target Tracking Device” of Japanese Patent Application Laid-Open No. 8-271617. In this method, processing efficiency is improved by processing detection data, a track that is a combination of the detection data, and a hypothesis that is a combination of the tracks for each cluster that does not share the detection data.
[0009]
FIG. 9 is an overall configuration diagram showing a conventional target tracking device disclosed in, for example, Japanese Patent Application Laid-Open No. 8-271617. FIG. 10 is a flowchart showing the operation of the conventional target tracking device shown in FIG. The operation of the conventional target tracking device will be described below with reference to FIGS. 9 and 10.
[0010]
First, in step S021 "Read next sampling data" in FIG. 10, the detection data selection unit 0 shown in FIG. 9 reads detection data at that time from the sensor.
[0011]
In the next step S022 “in-gate determination”, the detection data selection unit 0 checks the correlation between the gate of the existing track calculated by the gate calculation unit 7 and the detection data, and to which cluster the incoming detection data belongs. decide.
[0012]
Here, FIG. 11 and FIG. 12 will be described as an example. Assume that three samples are obtained by time t = 2 in a cluster and there are five detection data, and the state is as shown in FIG. At this time, the wake is expressed as a set of detection data,
T1: 10-20
T2: 10-21
T3: 11-20
T4: 11-21
T5: 20
T6: 21
It is.
[0013]
In addition, the hypothesis for adopting a wake is expressed as “ID: {wake set} reliability”,
H1: {T1, T4} rel1
H2: {T1, T6} rel2
H3: {T2, T5} rel3
H4: {T2} rel4
H5: {T5} rel5
Suppose that In step S021 "Read next sampling data", the detection data 30, 31, 32, 33 are entered, the detection data 30, 31 are the gates of the wakes of T1, T3, T5, and the detection data 32 is the gates of T2, T4. Suppose you enter. It is also assumed that the detection data 33 has not entered any wake gate.
[0014]
In the next step S023 “Creation of a wake and a new cluster”, a wake is created by the new cluster integration unit 1 based on the correlation between the gate and the detected data. There are three types of wakes to be created:
[0015]
(1) Update track: A track that updates the existing track by adding detection data that has entered the gate.
(2) New wake: A wake starting from the detected data at that time
(3) Memory track wake: A track that extends the specification of the motion based on the information that “there was no relevant detection data at the corresponding time” with respect to the existing track.
[0016]
In the example of FIG. 12, first, as an update track,
T11: 10-20-30
T12: 10-20-31
T21: 10-21-32
T31: 11-20-30
T32: 11-20-31
T41: 11-21-32
T51: 20-30
T52: 20-31
T61: 21-32
[0017]
As a new wake,
T71: 30
T72: 31
T73: 32
Is generated,
[0018]
As a memory track wake,
T81: 10-20
T82: 10-21
T83: 11-20
T84: 11-21
T85: 20
T86: 21
Is created.
[0019]
The wake is a temporary wake when it is just generated as a new wake, and then continues to be updated until it is recognized as the main wake. Here, the provisional wake is a wake at a stage where it is determined whether or not the target is being tracked, and the main wake is a wake that has been determined to be tracking the target.
[0020]
In step S023, for detection data that has no correlation with any cluster, a new cluster is generated starting from the detection data. In the example of FIG. 12, since the detection data 33 has no correlation with any wake, a new cluster based on this detection data is generated.
[0021]
Further, in this step S023, when an equivalence relationship described later is generated between a plurality of clusters by the detection data at this time, processing for integrating the clusters is performed.
[0022]
In the next step S024 “Generate wake correlation matrix for each cluster”, first, an in-gate determination matrix, which is a list of interpretation possibilities for detection data correlated with each cluster, is generated by the in-gate determination matrix calculation unit 2. The
[0023]
Each row of the in-gate determination matrix indicates detection data correlated with the cluster at the current time. In this example, the first row corresponds to the detection data 30, the second row corresponds to the detection data 31, and the third row corresponds to the detection data 32. Regarding the columns, the first column indicates unnecessary signals, the second to seventh columns indicate existing tracks (T1 to T6 in this example), and the last three columns are new tracks. Indicates. Since all the detection data has a possibility of being an unnecessary signal, all the elements in the first column are 1. For the second column, the detection data correlated with the existing track T1 is 30, 31, so the first and second rows are 1 and the third row is 0. For the last three columns, the three new wakes can only correspond to their starting points, so they are unit matrices.
[0024]
Next, the wake correlation matrix calculation unit 3 creates a wake correlation matrix from the in-gate determination matrix. In the wake correlation matrix, the element of the in-gate determination matrix that is the same as the element that is 1 is always 1. In each row, the number of elements that are 1 is only one, and each column has 1 at most. However, the number of elements of 1 for the first column may be arbitrary. All the matrices that satisfy the condition. This is a combination of interpretations of the respective detection data that do not contradict each other. FIG. 14 shows some examples of the in-gate determination matrix and the wake correlation matrix (a part of the entire six) obtained in this example.
[0025]
In the next step S025 “Create hypothesis”, the hypothesis update unit 4 performs a process of updating the existing hypothesis with the wake correlation matrix. In updating each hypothesis, all the wake correlation matrices except for the wake correlation matrix assuming the existence of wakes other than the wake included in the hypothesis are used. For example, in hypothesis H1, T1 and T4 are referenced. In the wake correlation matrices (1) to (5), since (2) assumes the existence of the wake T2, it is not used for updating the hypothesis H1. When H1 is updated by (1), this indicates that all detection data at t = 3 are unnecessary signals.
H11: {T81, T84}
It becomes.
[0026]
If H1 is updated by (4),
H12: {T11, T41}
It becomes. When H1 is updated by (5),
H13: {T81, T72, T41}
It becomes. In this way, a new hypothesis is generated by updating an existing hypothesis with a combination of a possible hypothesis and a wake correlation matrix.
[0027]
In the next step S027 “Semi-optimization and cluster separation”, the hypothesis reduction unit 5 executes a semi-optimization process for reducing hypotheses by a method such as deleting hypotheses with low reliability and integrating similar hypotheses, The cluster separation unit 6 executes cluster separation processing associated therewith.
[0028]
There are various types of sub-optimization,
・ Set a threshold for reliability and delete all hypotheses with reliability less than that.
・ Set an upper limit for the number of hypotheses, leave only the set number of hypotheses in order of increasing reliability, and delete others.
-Integrate hypotheses with the same correlation contents of detection data for the past N times.
Such a method is known.
[0029]
Cluster separation is performed by the cluster separation unit 6 inspecting the sharing state (similarity relationship) of detection data between tracks as follows. In a cluster, as a general rule, all detection data constituting a wake that shares detection data must belong to the same cluster. For example
Ta: 11-20-30
Tb: 21-32
Tc: 20-32
When there are three wakes, Ta and Tc (Ta to Tc), Tb and Tc (Tb to Tc) share detection data, and any two wakes in the three wakes are connected to each other with a plurality of similar relationships. (In the case of Ta and Tb, Ta to Tc to Tb).
[0030]
Two tracks that can be connected by a plurality of such similar relationships are said to have an equivalence relationship. The wake groups that have the equivalence relation form the same cluster. Consider a case where Tc is deleted by semi-optimization in this situation. Since the remaining two tracks do not share detection data, a cluster can be formed independently of each other.
[0031]
In the next step S028 “main track determination”, a process is performed in which the wake determination unit 8 promotes a tentative track that is very likely to be tracking the target to the main track from the tentative track group. As a condition for determining that “the possibility of tracking the target is very high” is, for example,
・ The track is included in all remaining hypotheses.
Is mentioned.
[0032]
The above is the flow of data processing for one cycle. When one cycle ends, the next detection data is read and the processing of the next cycle is executed.
[0033]
Here, “tracking initialization in the tracking method” which is a problem of the present invention will be described. The target tracking function that estimates the track of the target trajectory using the detection data obtained from the sensor
1. Tracking initialization function to detect unknown targets
2. Tracking maintenance function that tracks known targets
Can be broadly classified.
[0034]
For the former, it is necessary to consider the possibility of a new target for each detection data, but for the latter, there is a difference that it is not necessary to consider the possibility of a new target. In tracking initialization, a wake of a new target is created at each time, and depending on the hypothesis, the wake may or may not be adopted, so the number of wakes differs for each hypothesis. Therefore, a hypothesis having a number of wakes different from the actual target number is maintained, so that there is a possibility that the detection of the target may be delayed or an erroneous wake that is contrary to reality may be detected.
[0035]
[Problems to be solved by the invention]
As described above, in the conventional tracking method, the number of tracks to be tracked (the number of tracks) differs according to the hypothesis, so that an erroneous track is detected as a target, or the target track is originally detected in a situation that should be detected. It may not be possible.
[0036]
The present invention has been made to solve the above-described problems.
・ Information on the number of detected detection data
・ Information such as “which target number of hypotheses tends to be highly reliable”
An object of the present invention is to obtain a target tracking device that estimates the number of targets using and implements an efficient and accurate tracking initialization process.
[0037]
[Means for Solving the Problems]
The target tracking device according to the present invention includes a gate calculation unit that calculates a possible area of detection data from an existing track included in each cluster, and detection data corresponding to a track for each cluster from the existing area and the entire detection data. Create a new cluster for the detection data selection part to be selected and detection data that cannot be matched with the track in the existing cluster, and also select the cluster corresponding to the case where there is detection data that can be associated with the track in multiple clusters A new cluster / integration unit for integration, an in-gate determination matrix calculation unit for creating an in-gate determination matrix based on the presence / absence of correlation between existing tracks and detection data, and the in-gate determination matrix In Based on this, a wake correlation matrix calculation unit that extracts all possible combinations of interpretation for each detection data and calculates a wake correlation matrix, and a hypothesis update that newly generates and updates a hypothesis by combining the existing hypothesis and the wake correlation matrix And low reliability from the above hypothesis Hypothetical Reduction Executes semi-optimization processing to reduce hypothesis group by A hypothesis reduction unit, a cluster separation unit that separates clusters based on whether or not detection data is shared between wakes, and a wake that is likely to be a target based on a predetermined condition from the wake Estimate the number of targets in the cluster Of the semi-optimization process A target tracking device comprising a sub-optimization target number estimation unit for determining content, wherein the sub-optimization for The target number estimation unit is a sampling-specific target number estimation unit that estimates the target number at each sampling time; The semi-optimization process Target number estimation timing determination unit for determining whether or not to determine the target number to be used for the determination, and target number determination for track number control for estimating the target number for track number control from the history of the target number for each sampling time And a track number limiting unit for hypothesis that deletes hypotheses based on the target number for track number control estimated by the target number determination unit for track number control.
[0038]
In addition, the target tracking device according to the present invention includes a gate calculation unit that calculates a possible area of detection data from an existing track included in each cluster, and a track-corresponding detection for each cluster from the existing area and the entire detection data. A new cluster is created for the detection data selection unit that selects data and detection data that cannot be matched with the tracks in the existing cluster, and there is detection data that can be associated with tracks in multiple clusters A cluster establishment / integration unit for integrating clusters, an in-gate determination matrix calculation unit for creating an in-gate determination matrix based on the presence / absence of correlation between existing tracks and detection data, and the in-gate determination matrix In All possible combinations of interpretation for each detection data Calculate the wake correlation matrix A wake correlation matrix calculation unit, a hypothesis update unit that newly generates and updates a hypothesis by combining the existing hypothesis and the wake correlation matrix, and a low reliability from the hypothesis Hypothetical Reduction Executes semi-optimization processing to reduce hypothesis group by A hypothesis reduction unit, a cluster separation unit that separates clusters based on whether or not detection data is shared between wakes, and a wake that is likely to be a target based on a predetermined condition from the wake A target tracking device, and a target tracking device that estimates the number of targets in a cluster and determines the number of tracks in a hypothesis to be generated. The estimating unit is a sampling-specific target number estimating unit that estimates the target number at each sampling time; The semi-optimization process Target number estimation timing determination unit for determining whether or not to determine the target number to be used for the determination, and target number determination for track number control for estimating the target number for track number control from the history of the target number for each sampling time And a wake correlation matrix limiting unit that determines the number of wakes that the update hypothesis should have for each wake correlation matrix.
[0039]
The target number estimation unit for each sampling calculates the target number for each sampling time using the volume of the cluster calculated from the track in the cluster and the number of detection data at the latest sampling time correlated within the cluster.
[0040]
The sampling target number estimation unit estimates the target number for each sampling time as the number of tracks for which the hypothesis is classified for each number of tracks and the sum of the reliability is the largest value.
[0041]
In addition, the sampling target number estimation unit estimates the target number at each sampling time as an average value by weighting the number of hypothetical tracks with reliability.
[0042]
The target number estimation time determination unit uses a difference between the estimated target number at the current sampling time and the estimated target number at the previous sampling time as a criterion for determining the target number estimation time.
[0043]
The target number estimation time determination unit uses the number of samples that have elapsed since the previous target number determination time as a criterion for determining the target number estimation time.
[0044]
Further, the track number control target number determination unit estimates the average value of the target number estimation values at each sampling time as the track number control target number.
[0045]
Further, the track number control target number determination unit estimates a smooth value of a filter having the target number estimated value history as an input as the track number control target number.
[0046]
In addition, the hypothesis reduction unit “integrates hypotheses with the same correlation contents of detection data for the past N times” when the progress of the main wake is unlikely to occur to a predetermined degree in the process of the wake determination unit. In the hypothesis reduction method, the parameter N is set smaller as time advances.
[0047]
DETAILED DESCRIPTION OF THE INVENTION
Embodiment 1 FIG.
FIG. 1 is an overall configuration diagram showing a target tracking apparatus according to Embodiment 1 of the present invention. FIG. 2 is a diagram showing in detail the quasi-optimization target number estimation unit of FIG. FIG. 3 is a flowchart showing the operation of the target tracking device shown in FIG. FIG. 4 is a flowchart showing in detail the sub-optimization target number estimation step of FIG.
[0048]
In FIG. 1, 0 is a detection data selection unit that selects detection data corresponding to a track for each cluster from the possible existence area and the entire detection data. 1 is a new cluster that creates a new cluster for detection data that cannot be matched with tracks in an existing cluster, and also integrates the corresponding clusters when there is detection data associated with tracks in multiple clusters. It is an integrated department. Reference numeral 2 denotes an in-gate determination matrix calculation unit that creates an in-gate determination matrix based on the presence / absence of correlation between existing wakes and detection data. 3 is the in-gate decision matrix In This is a wake correlation matrix calculation unit that extracts all possible combinations of interpretation for each detection data and calculates a wake correlation matrix.
[0049]
Reference numeral 4 denotes a hypothesis update unit that newly generates and updates a hypothesis by combining an existing hypothesis and the above-described track correlation matrix. 5 is unreliable due to many hypotheses Hypothetical Reduction Executes semi-optimization processing to reduce hypothesis group by This is a hypothesis reduction unit. Reference numeral 6 denotes a cluster separation unit that separates clusters based on the presence or absence of sharing of detection data between wakes. Reference numeral 7 denotes a gate calculation unit that calculates an area where detection data can exist from existing tracks included in each cluster. Reference numeral 8 denotes a wake determination unit that selects a wake having a high possibility of being a target based on a predetermined condition from the wake and sets it as a main wake.
[0050]
Further, 9 is a semi-optimization target number estimation unit that estimates the target number in the cluster and determines the processing contents of the semi-optimization. As shown in FIG. 2, the pre-optimization target number estimation unit 9 includes a sampling-specific target number estimation unit 91 that estimates the target number at each sampling time, and determines the target number used for the semi-optimization. Target number estimation time determination unit 92 for determining whether or not to perform, a track number control target number determination unit 93 for estimating a track number control target number from a history of target numbers at each sampling time, and a track number control A track number limit unit for hypothesis 94 that deletes hypotheses based on the track number control target number estimated by the target number determination unit.
[0051]
Next, the operation of the present embodiment will be described with reference to FIGS.
First, in step S001 “Read next sampling data”, the detection data selection unit 0 reads the detection data at that time from the sensor.
In the next step S002 “determination within gate”, the detection data selection unit 0 performs the inside / outside determination of the detection data at the gate of the existing wake and inspects the correlation state to determine the cluster to which the incoming detection data belongs.
[0052]
In the next step S003 “Creation of Wake and New Cluster”, the new cluster integration unit 1 creates a wake based on the correlation between the existing wake gate and the detected data. There are three types of wakes to be created:
(1) Update track: An update track created by adding the detection data that entered the gate to the existing track.
(2) New wake: A wake starting from the detected data at that time
(3) Memory track wake: An updated wake that extends motion specifications based on the information that “there was no correlated detection data at the corresponding time” for the existing track
[0053]
In step S003, for detection data that has no correlation with any cluster, a new cluster is generated starting from the detection data. Further, in this step S003, when an equivalence relationship occurs between a plurality of clusters due to the detection data at this time, processing for integrating the clusters is performed.
[0054]
In the next step S 004 “Wake correlation matrix generation for each cluster”, an in-gate determination matrix is first generated by the in-gate determination matrix calculation unit 2. The in-gate determination matrix is a table showing a possible correlation between detection data and a wake.
[0055]
Next, the wake correlation matrix calculation unit 3 creates a wake correlation matrix from the in-gate determination matrix. The wake correlation matrix is
The element in the in-gate determination matrix that is the same as the element that is 1 is always 1.
・ Each row has only 1 element.
・ Each row has at most one element. However, the first column may have any number of elements.
All the matrices that satisfy the condition. This is a combination of interpretations of the respective detection data that do not contradict each other.
[0056]
In the next step S005 “Create hypothesis”, the hypothesis update unit 4 executes a process of updating the existing hypothesis with the wake correlation matrix. Each hypothesis update uses all the wake correlation matrices, except for the wake correlation matrix, which assumes the existence of wakes other than the wakes contained in it, and combines the possible hypotheses with the wake correlation matrix. Update existing hypotheses and generate new hypotheses.
The processing so far is exactly the same as the conventional example.
[0057]
In the next step S006 “Semi-optimization target number estimation”, the semi-optimization target number estimation unit 9 performs target number estimation processing. The operation of the semi-optimization target number estimation unit 9 will be described with reference to FIGS. First, in step S0061 “Sampling target number calculation”, the sampling target number estimation unit 91 performs estimation calculation of the sampling target number based on the information regarding tracking at the current sampling time. The method is one of the following.
[0058]
(1) Estimation method using the number of input detection data
Sampling time t _k The number of detected data correlated to a certain cluster in m _k And The volume of the cluster is V _k , False alarm density β _FT, The detection probability is P _dk And From these parameters, the target number by sampling at the current sampling time is
[0059]
[Expression 1]

[0060]
Estimate as follows. Where β _FT , P _dk Is a parameter set in advance. In contrast, the cluster volume V _k Integrates all the tracks belonging to the cluster, and calculates an integrated prediction value and an integrated prediction error covariance according to the following method.
[0061]
There are M wakes in the cluster, _i (I = 1,..., M), each track reliability is β _i (I = 1,..., M). Wake T _i Let X be the prediction vector _{k, i} Underscore (-), prediction error covariance matrix P _{k, i} (-). At this time, the predicted value of the wake formed by integrating the M wakes is calculated by Equation (2), and the prediction error covariance matrix is calculated by Equation (3).
[0062]
[Expression 2]

[Equation 3]

[0063]
An integrated gate is obtained from the integrated prediction value and the integrated prediction error covariance matrix, and the volume of the cluster is approximated by the volume.
Here, integrated gate is
[0064]
[Expression 4]

[0065]
This is an area that can be taken by the Z underbar that satisfies the conditional expression (5) using the residual covariance matrix obtained by the above.
[0066]
[Equation 5]

[0067]
(2) Hypothesis estimation based on reliability for each number of tracks
Each hypothesis is classified by the number of tracks. The sum of hypothesis reliability is calculated for each classification, and the number of tracks in the classification with the largest sum is set as the target number.
[0068]
Hereinafter, an example will be described. It is assumed that the following five hypotheses (“ID: {Wake Group} reliability”) are generated.
H1: {T1} rel1
H2: {T1, T3, T6} rel2
H3: {T2, T5} rel3
H4: {T2, T3, T6} rel4
H5: {T5, T6} rel5
[0069]
Focusing on the number of tracks that the hypothesis refers to,
Track number 1 group: H1, reliability rel1
Wake number 2 group: H3, H5, reliability rel3 + rel5
Wake number 3 group: H2, H4, reliability rel2 + rel6
[0070]
Of these three groups, it is assumed that the reliability of the group based on the wake number 2 hypothesis is the highest. At this time, the estimated value of the target number for each sampling at the current sampling time is 2.
[0071]
(3) Estimation by weighted average based on the reliability of the number of tracks
Let p hypotheses in the cluster be H _i (I = 1,..., P), the reliability of each is β _i (I = 1, ..., p), and the number of tracks is t _i (I = 1,..., P). At this time, the estimated value of the target number for each sampling at the current sampling time is obtained by equation (6).
[0072]
[Formula 6]

[0073]
This is a weighted average based on the hypothetical reliability of the number of tracks.
[0074]
Next, the target number estimation time determination unit 92 determines whether the tracking situation satisfies a target number determination condition for the number of tracks. The condition for this determination is one of the following two.
(1) The difference between the estimated value of the target number by sampling at the current sampling time and the estimated value of the target number by sampling at the previous sampling time is smaller than a preset threshold value.
(2) When the number of samples that have elapsed since the previous target number determination time for track number control reaches a threshold value.
[0075]
If it is determined that the target number determination condition is satisfied, the track number control target number determination unit 93 performs step S0062 “target number determination for track number control” processing, and then the hypothetical track limit limiting unit 94 performs step S0063 “under hypothesis. Perform wake limit. If the target number determination condition is not satisfied, the process proceeds to the next step S007 “Semi-optimization and cluster integration” without doing anything.
[0076]
In step S0062 “target number determination for track number control”, the target number for track number control is calculated from the time series of the target number estimated values by sampling. This calculation method is one of the following two.
[0077]
(1) Calculation based on overall average
The target number estimate by sampling that is not used for the target number determination for track number control executed before this stage is N _i Let ^ (i = 1,... R). At this time, the target number is determined by the following equation (7).
[0078]
[Expression 7]

[0079]
(2) Calculation by filter smoothing
The smoothing value of the filter is used to input the history of the estimated number of target values for each sampling time from the tracking processing start time. The following is a calculation formula using an α filter.
[0080]
[Equation 8]

[Equation 9]

[0081]
Where N _sk Is the sampling time t _k Smooth value of the target number in N, N _pk Is the predicted number of targets, N _k ^ Is input to the filter, sampling time t _k It is the target number estimated value according to sampling in. Α _k Is a parameter.
[0082]
In Step S0063 “Limit number of tracks in hypothesis”, only the hypotheses whose track number matches the target number for track number control determined in the previous step from the hypothesis group at the current sampling time by the hypothesis track limit unit 94. Remove all hypotheses that do not match.
[0083]
In the next step S007 "Semi-optimization and cluster separation", the hypothesis reduction unit 5 executes a semi-optimization process for reducing hypotheses by deleting hypotheses with low reliability and integrating similar hypotheses, and further, a cluster separation unit 6 executes the cluster separation process associated therewith.
[0084]
In this semi-optimization process, when it is difficult to promote the main track in the next step S008 “main track determination”, the hypothesis “integrates hypotheses with the same correlation contents of the detection data for the past N times”. In the reduction, the parameter N may be set smaller as the time advances. By performing hypothesis reduction with this setting, the main track can be detected more quickly in the “main track determination”.
[0085]
In the next step S008 “main track determination”, a process is performed in which the wake determination unit 8 promotes a tentative track that is highly likely to be tracking the target from the tentative track group to the main track.
[0086]
Thus, in the present embodiment, the number of targets existing in the cluster is estimated based on the number of detection data or the number of tracks in the hypothesis, and is used to limit the number of tracks in the hypothesis. The identification speeds up, so that the target track can be detected efficiently. Moreover, since a hypothesis having an incorrect number of tracks can be excluded, the target track can be detected more accurately.
[0087]
Embodiment 2. FIG.
FIG. 5 is an overall configuration diagram showing a target tracking device according to Embodiment 2 of the present invention. FIG. 6 is a diagram showing in detail the hypothesis generation target number estimation unit of FIG. FIG. 7 is a flowchart showing the operation of the hypothesis generation target estimation unit shown in FIG. FIG. 8 is a flowchart showing in detail the hypothesis generation target estimation step of FIG.
[0088]
As shown in FIG. 5, the target tracking device according to the present embodiment includes a hypothesis generation target number estimation unit 10 that estimates the target number in the cluster and determines the number of tracks in the hypothesis to be generated. Further, as shown in FIG. 6, the hypothesis generation target number estimation unit 10 determines a target number to be used for semi-optimization and a target number estimation unit 101 for each sampling that estimates the target number at each sampling time. A target number estimation time determination unit 102 for determining whether or not to perform, a track number control target number determination unit 103 for estimating a track number control target number from a history of target numbers for each sampling time, and a track correlation matrix The wake correlation matrix limiting unit 104 determines the number of wakes that the update hypothesis should have.
Other configurations are the same as those in the first embodiment.
[0089]
Next, the operation of the present embodiment will be described with reference to FIGS.
First, in step S011, “reading of next sampling data”, the detection data selection unit 0 reads detection data at that time from the sensor.
[0090]
In the next step S012 “in-gate determination”, the detection data selection unit 0 performs an internal / external determination of the detection data at the gate of the existing track to inspect the correlation state and determine a cluster to which the incoming detection data belongs.
[0091]
In the next step S013 “Creation of a wake and a new cluster”, a wake is created by the new cluster integration unit 1 based on the correlation between the existing wake gate and the detection data. There are three types of tracks to be created: update tracks, new tracks, and memory track tracks.
[0092]
In step S013, for the detection data that has no correlation with any cluster, a new cluster starting from the detection data is generated. Further, in this step S013, when an equivalence relationship occurs between a plurality of clusters due to the detection data at this time, processing for integrating the clusters is performed.
[0093]
In the next step S014 “Wake correlation matrix generation for each cluster”, first, an in-gate determination matrix is generated by the in-gate determination matrix calculation unit 2. The in-gate determination matrix is a table showing a possible correlation between detection data and a wake.
[0094]
Next, the wake correlation matrix calculation unit 3 creates a wake correlation matrix from the in-gate determination matrix. The wake correlation matrix is
The element in the in-gate determination matrix that is the same as the element that is 1 is always 1.
-Only one element can be 1 in each row.
・ Each row has at most one element. However, the first column may have any number of elements.
All the matrices that satisfy the condition. This is a combination of interpretations of the respective detection data that do not contradict each other.
The processing so far is substantially the same as in the first embodiment.
[0095]
In the next step S015 “Hypothesis generation target number estimation”, the hypothesis generation target number estimation unit 10 performs target number estimation processing. The operation of the hypothesis generation target number estimation unit 10 will be described with reference to FIGS. First, in step S0151, “target number calculation by sampling”, the target number estimation unit 101 by sampling performs an estimation calculation of the target number by sampling based on the information regarding tracking at the latest sampling time. The method will be described below. This is the same method as “(1) Estimation method using the number of input detection data” in the step of the same name in the first embodiment.
[0096]
Sampling time t _k The number of detected data correlated to a certain cluster in m _k And The volume of the cluster is V _k , False alarm density β _FT , P is the detection probability _dk And From these parameters, the estimated target number of the current sampling time is
[0097]
[Expression 10]

[0098]
Estimate as follows. Where β _FT, P _dk Is a parameter set in advance. In contrast, the cluster volume V _k Integrates the wake groups belonging to the cluster, calculates the integrated prediction value and the integrated prediction error covariance, calculates the integrated gate from these integrated prediction values and the integrated prediction error covariance matrix, and calculates the volume of the cluster by its volume. Approximate.
[0099]
Next, the target number estimation time determination unit 102 determines whether the tracking status at the current time satisfies the target number determination condition. The condition for this determination is one of the following two. This is the same method as (1) and (2) in the determination step of the same name in the first embodiment.
[0100]
(1) The difference between the estimated value of the target number by sampling at the current sampling time and the estimated value of the target number by sampling at the previous sampling time is smaller than a preset threshold value.
(2) When the number of samples that have elapsed since the previous target number determination time for track number control reaches a threshold value.
[0101]
When it is determined that the target number determination condition is satisfied, the track number control target number determination unit 103 performs step S0152 “determination of track number control target number” processing, and then the wake correlation matrix restriction unit 104 performs step S0153. Perform “Wake correlation matrix restriction”. If the target number determination condition is not satisfied, the process proceeds to the next step S016 “Generate hypothesis” without doing anything.
[0102]
In step S0152 “target number determination for track number control”, the target number for track number control is calculated from the target number estimated value calculated at each sampling time. This calculation method is one of the following two. This is exactly the same as (1) and (2) in the steps of the same name in the first embodiment.
[0103]
(1) Calculation based on overall average
The target number estimate by sampling that is not used for the target number determination for track number control executed before this stage is N _i Let ^ (i = 1,...). At this time, the target number is determined by the following equation (7).
[0104]
[Expression 11]

[0105]
(2) Calculation by filter smoothing
It is calculated by the smooth value of the filter using as input the time series of the sampling target number estimation values for each sampling time from the tracking processing start time.
[0106]
In step S0153 “track correlation matrix restriction”, the necessary number of existing tracks is registered for all the track correlation matrices generated in the previous step S014 “generation of track correlation matrix for each cluster” by the wake correlation matrix restriction unit 104. . The required number of existing tracks is
[0107]
Estimated number of targets for track number control-Number of elements for which "1" is set for the new target part in the track correlation matrix
[0108]
It is. If the wake correlation matrix is composed of n rows and m columns, the new target portion of the wake correlation matrix is the portion of the mn + 1st column to the mth column. It is assumed that the target estimated value is determined to be 2. In the example of the wake correlation matrix in FIG. 14, the required number of existing wakes is as follows.
[0109]
(1) → 2
(2) → 1
(3) → 1
(4) → 2
(5) → 1
(6) → -1
[0110]
The required number of existing wakes calculated here is the number of wakes that the existing hypothesis that can make an update hypothesis in combination with the wake correlation matrix should have. Further, when the required number of existing tracks becomes a value less than 0, the track correlation matrix is deleted. In the example of FIG. 14, the track correlation matrix (6) is deleted.
[0111]
In the next step S016 “Create hypothesis”, the hypothesis update unit 4 executes a process of updating the existing hypothesis with the wake correlation matrix. Each hypothesis update uses all the wake correlation matrices, except for the wake correlation matrix, which assumes the existence of wakes other than the wakes contained in it, and combines the possible hypotheses with the wake correlation matrix. Update existing hypotheses and generate new hypotheses.
[0112]
If the target number for track number control has been estimated in the process of step S0152 at the current sampling time, the existing hypothesis can be updated in this step because the existing track required number of the combined track correlation matrix and the existing track number are existing. Only when the number of tracks in the hypothesis matches. The necessary number of existing tracks is calculated and registered for each track correlation matrix in step S0153 “Track correlation matrix restriction”. For the wake correlation matrix of FIG.
[0113]
From (1), only existing hypotheses with 2 existing tracks can be combined.
From (2), only existing hypotheses with 1 existing track can be combined.
From (3), only existing hypotheses with 1 existing track can be combined.
From (4), only existing hypotheses with 2 existing tracks can be combined.
From (5), only existing hypotheses with 1 existing track can be combined.
It becomes.
If the number of existing wakes is larger than the target number estimate for wake number control, no update hypothesis is created from this existing hypothesis.
[0114]
In the next step S017 “Semi-optimization and cluster separation”, the hypothesis reduction unit 5 executes a semi-optimization process for reducing hypotheses by deleting hypotheses with low reliability and integrating similar hypotheses, and further, separating clusters. The unit 6 executes cluster separation processing associated therewith.
[0115]
In the next step S018 “main track determination”, a process is performed in which the wake determination unit 8 promotes a tentative track that is very likely to be tracking the target from the tentative track group to the main track.
[0116]
As described above, in the present embodiment, the number of targets existing in the cluster is determined based on the number of detection data, and is used to limit the number of tracks at the time of hypothesis generation. This makes it possible to efficiently detect the target track. Moreover, since a hypothesis having an incorrect number of tracks can be excluded, the target track can be detected more accurately.
[0117]
【The invention's effect】
The target tracking device according to the present invention includes a gate calculation unit that calculates a possible area of detection data from an existing track included in each cluster, and detection data corresponding to a track for each cluster from the existing area and the entire detection data. Create a new cluster for the detection data selection part to be selected and detection data that cannot be matched with the track in the existing cluster, and also select the cluster corresponding to the case where there is detection data that can be associated with the track in multiple clusters A new cluster / integration unit for integration, an in-gate determination matrix calculation unit for creating an in-gate determination matrix based on the presence / absence of correlation between existing tracks and detection data, and the in-gate determination matrix In Based on this, a wake correlation matrix calculation unit that extracts all possible combinations of interpretation for each detection data and calculates a wake correlation matrix, and a hypothesis update that newly generates and updates a hypothesis by combining the existing hypothesis and the wake correlation matrix And low reliability from the above hypothesis Hypothetical Reduction Executes semi-optimization processing to reduce hypothesis group by A hypothesis reduction unit, a cluster separation unit that separates clusters based on whether or not detection data is shared between wakes, and a wake that is likely to be a target based on a predetermined condition from the wake Estimate the number of targets in the cluster Of the semi-optimization process A target tracking device comprising a sub-optimization target number estimation unit for determining content, wherein the sub-optimization for The target number estimation unit is a sampling-specific target number estimation unit that estimates the target number at each sampling time; The semi-optimization process Target number estimation timing determination unit for determining whether or not to determine the target number to be used for the determination, and target number determination for track number control for estimating the target number for track number control from the history of the target number for each sampling time And a track number limiting unit for hypothesis that deletes hypotheses based on the target number for track number control estimated by the target number determination unit for track number control. For this reason, the number of targets existing in the cluster is estimated based on the number of detected data or the number of tracks in the hypothesis, and is used to limit the number of tracks in the hypothesis. It is possible to efficiently detect the wake. Moreover, since a hypothesis having an incorrect number of tracks can be excluded, the target track can be detected more accurately.
[0118]
In addition, the target tracking device according to the present invention includes a gate calculation unit that calculates a possible area of detection data from an existing track included in each cluster, and a track-corresponding detection for each cluster from the existing area and the entire detection data. A new cluster is created for the detection data selection unit that selects data and detection data that cannot be matched with the tracks in the existing cluster, and there is detection data that can be associated with tracks in multiple clusters A cluster establishment / integration unit for integrating clusters, an in-gate determination matrix calculation unit for creating an in-gate determination matrix based on the presence / absence of correlation between existing tracks and detection data, and the in-gate determination matrix In All possible combinations of interpretation for each detection data Calculate the wake correlation matrix A wake correlation matrix calculation unit, a hypothesis update unit that newly generates and updates a hypothesis by combining the existing hypothesis and the wake correlation matrix, and a low reliability from the hypothesis Hypothetical Reduction Executes semi-optimization processing to reduce hypothesis group by A hypothesis reduction unit, a cluster separation unit that separates clusters based on whether or not detection data is shared between wakes, and a wake that is likely to be a target based on a predetermined condition from the wake A target tracking device, and a target tracking device that estimates the number of targets in a cluster and determines the number of tracks in a hypothesis to be generated. The estimating unit is a sampling-specific target number estimating unit that estimates the target number at each sampling time; The semi-optimization process Target number estimation timing determination unit for determining whether or not to determine the target number to be used for the determination, and target number determination for track number control for estimating the target number for track number control from the history of the target number for each sampling time And a wake correlation matrix limiting unit that determines the number of wakes that the update hypothesis should have for each wake correlation matrix. For this reason, the number of targets existing in the cluster is determined based on the number of detection data and used to limit the number of tracks when generating hypotheses. Can be performed. Moreover, since a hypothesis having an incorrect number of tracks can be excluded, the target track can be detected more accurately.
[0119]
The target number estimation unit for each sampling calculates the target number for each sampling time using the volume of the cluster calculated from the track in the cluster and the number of detection data at the latest sampling time correlated within the cluster. That is, since the target number for each sampling is calculated using the volume of the cluster calculated from the track in the cluster and the number of detection data correlated in the cluster, it is possible to accurately estimate the target number reflecting the observation situation.
[0120]
The sampling target number estimation unit estimates the target number for each sampling time as the number of tracks for which the hypothesis is classified for each number of tracks and the sum of the reliability is the largest value. In other words, since the number of tracks with the largest sum of reliability is estimated as the target number by sampling, after classifying hypotheses for each number of tracks, target number estimation that reflects the contents of the hypothesis maintained by the tracking device is performed. It becomes possible.
[0121]
In addition, the sampling target number estimation unit estimates the target number at each sampling time as an average value by weighting the number of hypothetical tracks with reliability. That is, since the number of tracks of hypotheses is weighted with reliability and the average value is estimated as the target number by sampling, it is possible to accurately estimate the target number reflecting the contents of the hypothesis maintained by the tracking device.
[0122]
The target number estimation time determination unit uses a difference between the estimated target number at the current sampling time and the estimated target number at the previous sampling time as a criterion for determining the target number estimation time. In other words, since the difference between the estimated target number at the current sampling time and the estimated target number at the previous sampling time is used as a criterion for determining the target number, it is possible to determine the target number when the target number estimation is stable, Accurate target number can be determined.
[0123]
The target number estimation time determination unit uses the number of samples that have elapsed since the previous target number determination time as a criterion for determining the target number estimation time. That is, since the number of samples that have elapsed since the previous target number determination time is used as a criterion for determining the target number determination time, the target number determination is performed periodically, and the target number cannot be determined after any time.
[0124]
Further, the track number control target number determination unit estimates the average value of the target number estimation values at each sampling time as the track number control target number. That is, since the average value of the target number estimation values at each sampling time is estimated as the target number, it is possible to accurately estimate the target number reflecting the target number at each sampling time.
[0125]
Further, the track number control target number determination unit estimates a smooth value of a filter having the target number estimated value history as an input as the track number control target number. That is, since the estimation is performed as a smooth value of the filter using the history of the target number estimation value as an input, the target number at each sampling time is considered as a time series, and more accurate target number estimation is possible.
[0126]
In addition, the hypothesis reduction unit “integrates hypotheses with the same correlation contents of detection data for the past N times” when the progress of the main wake is unlikely to occur to a predetermined degree in the process of the wake determination unit. In the hypothesis reduction method, the parameter N is set smaller as time advances. That is, in the hypothesis reduction method of “integrating hypotheses with the same correlation contents of detection data in the past N times”, the target wake uses a wake establishment criterion in order to set the parameter N smaller as time advances. It becomes easy to satisfy, and the target track can be detected early.
[Brief description of the drawings]
FIG. 1 is an overall configuration diagram showing a target tracking device according to a first embodiment of the present invention.
FIG. 2 is a diagram illustrating in detail a quasi-optimization target number estimation unit in FIG. 1;
FIG. 3 is a flowchart showing an operation of the target tracking device shown in FIG. 1;
FIG. 4 is a flowchart showing in detail a sub-optimization target number estimation step of FIG. 3;
FIG. 5 is an overall configuration diagram showing a target tracking device according to a second embodiment of the present invention.
6 is a diagram illustrating in detail a hypothesis generation target number estimation unit in FIG. 5;
7 is a flowchart showing an operation of a hypothesis generation target estimation unit shown in FIG. 5;
FIG. 8 is a flowchart showing in detail the hypothesis generation target estimation step of FIG. 7;
FIG. 9 is an overall configuration diagram showing a conventional target tracking device.
10 is a flowchart showing the operation of the conventional target tracking device shown in FIG.
FIG. 11 is an explanatory diagram for explaining a correlation degree of detection data sampled at each time.
FIG. 12 is an explanatory diagram for explaining a correlation degree of detection data sampled at each time;
FIG. 13 is an explanatory diagram showing an example of a correlation between a wake and detection data at each time.
FIG. 14 is a diagram illustrating an example of an in-gate determination matrix and a wake correlation matrix (a part of the entire six) obtained therefrom.
[Explanation of symbols]
0 detection data selection unit, 1 cluster establishment / integration unit, 2 in-gate determination matrix calculation unit, 3 wake correlation matrix calculation unit, 4 hypothesis update unit, 5 hypothesis reduction unit, 6 cluster separation unit, 7 gate calculation unit, 8 track Decision unit, 9 sub-optimization target number estimation unit, 91 sampling target number estimation unit, 92 target number estimation time determination unit, 93 track number control target number determination unit, 94 hypothesis track number limit unit, 10 hypothesis generation Target number estimation unit, 101 sampling target number estimation unit, 102 target number estimation time determination unit, 103 track number control target number determination unit, 104 track correlation matrix restriction unit.

Claims (10)

A gate calculation unit for calculating a possible area of detection data from existing tracks included in each cluster;
A detection data selection unit for selecting detection data corresponding to a wake for each cluster from the existence possible area and the entire detection data;
A new cluster / integration unit that creates a new cluster for detection data that cannot be matched with tracks in existing clusters, and integrates the corresponding clusters when there is detection data associated with tracks in multiple clusters ,
An in-gate decision matrix calculation unit that creates an in-gate decision matrix from the presence or absence of correlation between existing wakes and detection data,
Based on the in-gate determination matrix , a wake correlation matrix calculation unit that extracts all possible combinations of interpretation for each detection data and calculates a wake correlation matrix;
A hypothesis updating unit that generates and updates a new hypothesis by combining the existing hypothesis and the wake correlation matrix;
A hypothesis reduction unit that performs a sub-optimization process for reducing hypotheses by reducing hypotheses with low reliability from the hypothesis;
A cluster separation unit that separates clusters based on the presence or absence of sharing of detection data between wakes;
A wake determination unit that selects a wake that is highly likely to be a target based on predetermined conditions from the wake and sets it as a main wake,
A target tracking device comprising: a target number estimation unit for semi-optimization that estimates the target number in a cluster and determines the content of the semi-optimization process;
The quasi optimized for target-count-estimation unit,
A target number estimator by sampling for estimating the target number at each sampling time;
A target number estimation time determination unit that determines whether or not to determine the target number used for the semi-optimization process ;
A target number determination unit for track number control for estimating the target number for track number control from the history of the target number for each sampling time;
A target tracking device comprising: a track number limit unit for hypothesis that deletes a hypothesis based on the target number for track number control estimated by the target number determination unit for track number control. A gate calculation unit for calculating a possible area of detection data from existing tracks included in each cluster;
A detection data selection unit for selecting detection data corresponding to a wake for each cluster from the existence possible area and the entire detection data;
A new cluster / integration unit that creates a new cluster for detection data that cannot be matched with tracks in existing clusters, and integrates the corresponding clusters when there is detection data associated with tracks in multiple clusters ,
An in-gate decision matrix calculation unit that creates an in-gate decision matrix from the presence or absence of correlation between existing wakes and detection data,
Based on the in-gate determination matrix , a wake correlation matrix calculation unit that extracts all possible combinations of interpretation for each detection data and calculates a wake correlation matrix;
A hypothesis updating unit that generates and updates a new hypothesis by combining the existing hypothesis and the wake correlation matrix;
A hypothesis reduction unit that performs a sub-optimization process for reducing hypotheses by reducing hypotheses with low reliability from the hypothesis;
A cluster separation unit that separates clusters based on the presence or absence of sharing of detection data between wakes;
A wake determination unit that selects a wake that is highly likely to be a target based on predetermined conditions from the wake and sets it as a main wake,
A target tracking device comprising a target number estimation unit for hypothesis generation that estimates the target number in a cluster and determines the number of wakes in a hypothesis to be generated,
The hypothesis generation target number estimation unit includes:
A target number estimator by sampling for estimating the target number at each sampling time;
A target number estimation time determination unit that determines whether or not to determine the target number used for the semi-optimization process ;
A target number determination unit for track number control for estimating the target number for track number control from the history of the target number for each sampling time;
A track tracking matrix limiting unit that determines the number of tracks that the update hypothesis should have for each track correlation matrix. The sampling target number estimation unit calculates the target number for each sampling time by using the volume of the cluster calculated from the track in the cluster and the number of detection data at the latest sampling time correlated in the cluster. The target tracking device according to claim 1 or 2. The target number estimation unit by sampling estimates the target number at each sampling time as a number of tracks having the highest sum of reliability by classifying hypotheses for each number of tracks. The target tracking device according to 1 or 2.   3. The target tracking according to claim 1, wherein the sampling-specific target number estimation unit estimates the target number at each sampling time as an average value by weighting the number of hypothetical tracks with reliability. apparatus. 3. The target according to claim 1, wherein the target number estimation time determination unit uses a difference between the estimated target number at the current sampling time and the estimated target number at the previous sampling time as a determination criterion for the target number estimation time. Tracking device. The target tracking apparatus according to claim 1, wherein the target number estimation time determination unit uses the number of samples that have elapsed since the previous target number determination time as a determination criterion for the target number estimation time. The target tracking device according to claim 1, wherein the track number control target number determination unit estimates an average value of target number estimation values at each sampling time as the track number control target number. 3. The target according to claim 1, wherein the track number control target number determination unit estimates a smooth value of a filter having a history of target number estimation values as an input as the track number control target number. Tracking device. The hypothesis reduction unit, when the process of the track determination unit is in a situation where the promotion of the main track is unlikely to occur to a predetermined degree, “integrates hypotheses with the same correlation contents of detection data for the past N times” In the hypothesis reduction method, the target tracking device according to claim 1 or 2, wherein the parameter N is set to be smaller as time advances.

Images