JP4623402B2 - Fire detection equipment - Google Patents

Description

【００４７】
としたとき、Ｓ／σ²（σ²は母集団＝ノイズの分散）は自由度（Ｎ−１）のχ²分布に従う（Ｎ＝標本データ数）。よってＳ／σ²を検定統計量としてχ²片側検定することで、ノイズを除去する。
【００４８】
すなわち、図７に示すように、検定統計量Ｓ／σ²が自由度２９のχ²分布に従うとき、検定統計量Ｓ／σ²が例えば５２以上のときは、矢印Ａで示すように、ノイズ以外と判定し、５２未満のときは矢印Ｂで示すように、ノイズと判定する。
【００４９】
図３に戻って、前記のように、ノイズ判定部１１は、フレームメモリ１０に記録されている各ピクセル毎の一定時間にわたる輝度の時間変動を入力し、正規分布にモデル化したノイズと比較し、ピクセルの輝度変動がノイズによって生じたものかを判定する。
【００５０】
次に、ゆらぎと物体の区別について説明する。
【００５１】
火災などによるゆらぎを安定に抽出するためには、物体の移動の影響を取り除かなくてはならない。移動物体を取り除くため、ある区間においてデータのバラツキが均一かどうかを検証する。これは、物体が移動した場合には、ゆらぎとは異なり均一でない輝度の変動が生ずるためである。
【００５２】
図８（Ａ）に示すように、ゆらぎの場合には、フレームに対する炎の輝度変化は、１３０〜１６０の内にあり、ばらつきがほぼ均一である。これに対して、図８（Ｂ）に示すように、移動物体の場合には、フレームに対する物体通過時の輝度変化は、５０から２００近くまで大きく変化し、バラツキが均一でなくなる。
【００５３】
ここで、各データ点を均一の重さの重りに見立て、注目している区間を円形に配置した時の重心の位置に注目する。もし、バラツキが均一であるなら、重心は中心付近に集まるはずである。しかし、輝度の変動が移動物体により生じた場合には、均一でない分布が生じるため、重心は円の中心から大きく外れる。そのため、この重心位置が円の中心からどれだけ外れているかを測定することにより、移動物体とゆらぎを区別することができる。
【００５４】
図９（Ａ）に示すように、ゆらぎの場合には、時間変化に対して輝度のバラツキは均一となっており、図９（Ｂ）に示すように、注目している区間の時間軸を円状に配置し、輝度が成す重心位置Ｏを求めると、重心位置Ｏは円Ｅの中心付近になる。一方、図９（Ｃ）に示すように、物体が移動物体の場合には、時間変化に対する輝度は大きく変化し、バラツキが大きい。図９（Ｄ）に示すように、注目している区間の時間軸を円状に配置し、輝度が成す重心位置Ｏを求めると、重心位置Ｏは円Ｅの中心から大きく外れる。
【００５５】
これを図１０（Ａ）に示すように、輝度ｒ１〜ｒ４のデータがＤ１〜Ｄ４が４個の場合について説明する。
【００５６】
図１０（Ｂ）に示すように、時間軸を円状に配置し、円Ｅの中心Ｍから垂直な輝度ｒ１のデータＤ１、９０度の右側に輝度ｒ２のデータＤ２、１８０度の下側に輝度ｒ３のデータＤ３、９０度左側に輝度ｒ４のデータＤ４を配置すると、重心位置Ｏは円Ｅの中心Ｍから少し下側にくる。
【００５７】
このように、測定されたデータがゆらぎの場合、移動物体とは異なり、一定区間におけるデータのバラツキが均一であるので、その均一性を用いて検証する。
【００５８】
注目しているデータ区間を円形に配置し、各データ点を重りに見立て、その重りが成す重心位置によってゆらぎかどうかを判定する。
【００５９】
データを円形に配置した時の重心座標を（Ｘ_m，Ｙ_m）とする。すると、
【００６０】
【数２】

【００６１】
となる。ここで、Ｖ₀，Ｖ₁…は各点における輝度を表している。また、Ｎは評価に用いるデータの個数である。
【００６２】
今考えている点が「ゆらぎ」に属していると仮定する。ここでゆらぎにおける各輝度、Ｖ₀，Ｖ₁…は正規分布、Ｎd（μ，σ²）に従うものとする。
【００６３】
すると、Ｘ_mは次の正規分布に従うことが分かる。
【００６４】
【数３】

【００６５】
従って次のように変換することで、Ｘ_mを平均値＝０、標準偏差＝１の正規分布に正準化することができる。
【００６６】
【数４】

【００６７】
同様にＹ_mを変換したｙ_mを求め、その２乗和Ｒを求めると、Ｒは自由度の２のχ²分布に従う。
【００６８】
【数５】

【００６９】
Ｒを検定統計量として有意水準αによって、「今考えている点はゆらぎである」という仮定を検定することで、ゆらぎかどうかの判定を行う。
【００７０】
図１１はゆらぎの検定統計量Ｒと確率密度関係を示すグラフである。
【００７１】
図１１において、検定統計量Ｒは自由度２のχ²分布に従う。この例では、検定統計量Ｒが例えば９以上のときは矢印Ｇで示すように移動物体であると判定し、９未満のときは矢印Ｈで示すように、ゆらぎであると判定する。
【００７２】
図３に戻って、移動物体判定部１２は、前述したように、時間軸を円状に配置し、輝度が成す重心位置を求め、その重心位置により、ピクセルの輝度変動が物体の移動によって生じたものかゆらぎによるものかを判定する。
【００７３】
移動物体判定部１２は、ウェイトテーブル（テーブル）１４を有し、（式１）において重心位置（Ｘ_m，Ｙ_m）を計算する際、係数部分については、予め値を計算しておき、ウェイトテーブル１４に格納しておく。図１２（Ａ）に示すようにウェイトテーブル１４には、Ｘ座標用として、データcos(2π/N×0)、データcos(2π/N×1)、データcos(2π/N×2)・・・がそれぞれ格納され、図１２（Ｂ）に示すようにウェイトテーブル１４にはＹ座標用としてデータsin(2π/N×0)、データsin(2π/N×1)、データsin(2π/N×2)・・・がそれぞれ格納される。
【００７４】
このように、（式１）において（Ｘ_m，Ｙ_m）を計算する際、係数部分については、事前に値を決めることができるので、係数部分を予め計算しておきウェイトテーブル１４に記録しておく。（Ｘ_m，Ｙ_m）を計算する際に係数を改めて計算するのでなく、テーブル１４から値を参照することによって演算量を低減することができるようにしている。
【００７５】
移動物体判定部１２の出力は、ゆらぎ判定部１３に送られ、また、ノイズ判定部１１の出力もゆらぎ判定部１３に送られる。ゆらぎ判定部１３は、ノイズ判定部１１の出力と移動物体判定部１２の出力を統合して、ノイズでなく、かつ、移動物体でもない領域のみをゆらぎ領域として抽出する。
【００７６】
２値化処理部８は、赤外線カメラの場合、その監視カメラ１から入力画像をＡＤ変換部６で輝度値に変換されたデータを入力し、物体の高温部のみを抽出する。
【００７７】
すなわち、２値化処理部８は、監視カメラ１からの入力画像より、輝度値が一定の値以上である領域を取り出すことで画像を２値化し、その２値化された領域とゆらぎ抽出部で検出されたゆらぎ領域の共通部分を火災領域として抽出する。
【００７８】
２値化で抽出する輝度の範囲は、炎を検出する場合には炎によって生じる輝度の範囲を選択し、煙を抽出する場合には、煙によって生じる輝度の範囲を選択する。
【００７９】
図１３（Ａ）に示すように、Ｓで示すような輝度信号があるとき、しきい値Ｔを超える領域Ｘを取り出し、図１３（Ｂ）に示すように２値化信号Ｙとする。
【００８０】
また、２値化処理部８は、可視カメラの場合、その監視カメラ１からの入力画像の色成分により、特定の領域のみを抽出する。
【００８１】
すなわち、２値化処理部８は、監視カメラ１からの入力画像より、色が予め設定された範囲内にある領域のみを取り出すことで画像を２値化し、その２値化された領域とゆらぎ抽出部で検出されたゆらぎ領域の共通部分を火災領域として抽出する。
【００８２】
２値化で抽出する色の範囲は炎を抽出する場合には炎が生成する色範囲を選択し、煙を抽出する場合には煙によって生じる色範囲を選択する。
【００８３】
なお、この２値化の処理も、図１３に示したような方法で行う。
【００８４】
２値化処理部８の出力は、火災判定部９に送られ、また、ゆらぎ判定部１３の出力も火災判定部９に送られる。ゆらぎ判定部１３と２値化処理部８から出力された結果は火災判定部９によって最終判定が行われる。まず、２値化処理部８の結果が火災の条件を満たしていることを確認し、かつ、ゆらぎ判定部１３でゆらぎであると判定されたピクセルを火災候補領域とする。この火災候補領域が一定の面積以上となった場合、火災が発生したと判断して警報を発する。警報装置５ではベルを鳴動して火災の発生を報知し、モニター装置４には火災の発生が表示される。
【００８５】
図１４はノイズ判定部１１の処理を示すフローチャートである。
【００８６】
図１４において、まず、ステップＳ１でフレームメモリ１０のデータ（輝度値）から各ピクセル毎に平方和Ｓを計算する。
【００８７】
【数６】

【００８８】
次に、ステップＳ２で検定統計量Ｓ／σ²が自由度（Ｎ−１）のχ²分布に行うことからその有意確率Ｐを求める。σ²はノイズの分散、Ｎは標本データ数である。
【００８９】
図１５に示すように、ＭＥで示す面積（＝確率）が有意確率Ｐを示す。例えば、矢印Ｚで示すように、現在測定された検定統計量Ｓ／σ²の値が例えば３５のときは、有意確率Ｐは２０％となる。図１６は検定統計量Ｓ／σ²と有意確率Ｐの関係を示す。例えば、検定統計量Ｓ／σ²が１，５，１０では有意確率Ｐはすべて１００％、２０では８９％、３０では４１％、４０では８％、５０では１％である。
【００９０】
次に、ステップＳ３で求めた有意確率Ｐを予め設定してある有意水準αと比較する。
【００９１】
次に、有意確率Ｐが有意水準αより小さければステップＳ４で該当ピクセルは、ノイズでないと判定する。有意確率Ｐが有意水準αより大きいときは、ステップＳ５で該当ピクセルはノイズであると判定する。例えば、有意水準αを３％とすると、有意確率Ｐが３％以下のときはノイズでないと判定し、有意確率Ｐが３％を超えるときノイズであると判定する。
【００９２】
次に、ステップＳ６で全ピクセルについて処理が終了していないときは、ステップＳ１に戻り、処理が終了したときは、エンドとする。
【００９３】
図１７は移動物体判定部１２の処理を示すフローチャートである。
【００９４】
図１７において、まず、ステップＳ１１でフレームメモリ１０のデータ（輝度値）から各ピクセル毎に重心（Ｘ_m，Ｙ_m）を求める。重心（Ｘ_m，Ｙ_m）は、式１より求める。このとき、ウェイトテーブル１４に予め格納してある係数部分（図１２、参照）を取り出して計算し、演算量を低減する。
【００９５】
次に、ステップＳ１２で検定統計量Ｒが自由度２のχ²分布に従うことからその有意確率Ｐを求める。検定統計量Ｒは、式４より求める。なお、式４は、
【００９６】
【数７】

【００９７】
である。
【００９８】
検定統計量Ｒは、図１８に示すように、自由度２のχ²分布に従う。図１８において、ＭＥで示す領域が成す面積（＝確率）が有意確率Ｐを示す。今矢印Ｚで示すように、検定統計量Ｒの現在測定された値が例えば４であるとすると、有意確率Ｐは１４％になる。
【００９９】
図１９は検定統計量Ｒと有意確率Ｐの関係を示す。例えば、検定統計量Ｒが１のとき有意確率Ｐは６１％、３のときは２２％、５のときは８％、７のときは３％、９のときは１８％となる。
【０１００】
次に、ステップＳ１３では求めた有意確率Ｐを予め設定してある有意水準αと比較する。有意確率Ｐが有意水準αより小さいときは、ステップＳ１４で該当ピクセルは移動物体であると判定し、有意確率Ｐが有意水準αより大きいときは、ステップＳ１５で該当ピクセルは移動物体でないと判定する。有意水準αを例えば２％に設定すると、有意確率Ｐが２％以下のときは、該当ピクセルは移動物体であると判定し、有意確率が２％を超えるときは、該当ピクセルは移動物体でないと判定する。
【０１０１】
次に、ステップＳ１６で全ピクセルについてこの処理を終了したか判別し、終了していないときは、ステップＳ１１に戻り、全ピクセルについて処理を終了したときは、エンドとする。
【０１０２】
図２０はゆらぎ判定部１３の処理を示すフローチャートである。
【０１０３】
図２０において、まず、ステップＳ２１でノイズ判定部１１の出力から該当ピクセルはノイズか否かを判別する。ノイズのときは、ステップＳ２４に進み、ノイズでないときは、ステップＳ２２に進む。
【０１０４】
ステップＳ２２では移動物体判定部１２の出力から該当ピクセルは移動物体か否かを判別する。移動物体のときはステップＳ２４に進み、移動物体でないときは、ステップＳ２３に進む。ステップＳ２３では、該当ピクセルはノイズではなく、かつ、移動物体でないので、該当ピクセルはゆらぎであると判定する。
【０１０５】
ステップＳ２４では、該当ピクセルはノイズであるかまたは移動物体であるので、該当ピクセルはゆらぎでないと、判定する。
【０１０６】
そして、ステップＳ２５で全ピクセルのこの処理が終了したか否かを判別し、終了していないときは、ステップＳ２１に戻り、終了したときは、処理をエンドとする。こうして、ノイズでなく、かつ、移動物体でもない領域のみをゆらぎ領域として抽出する。
【０１０７】
図２１は２値化処理部８の処理を示すフローチャートである。ここでは、監視カメラ１が赤外線カメラであり、高温部を取り出す場合について説明する。
【０１０８】
図２１において、まず、ステップＳ３１で該当ピクセルの輝度値が一定の輝度のしきい値以上であるか否かを判別し、一定のしきい値未満のときは、ステップＳ３３に進み、一定のしきい値以上のときは、ステップＳ３２に進む。
【０１０９】
ステップＳ３２では該当ピクセルは火災候補領域であると判定する。すなわち、監視カメラ１からの入力画像より、輝度値が一定の値以上である領域を取り出すことで画像を２値化し、その２値化された領域を火災候補領域とする。
【０１１０】
２値化で抽出する輝度の範囲は、炎を検出する場合には炎によって生じる輝度の範囲を選択し、煙を抽出する場合には煙によって生じる輝度の範囲を選択する。該当ピクセルの輝度値が一定のしきい値未満のときは、該当ピクセルは火災でないと判定する。
【０１１１】
ステップＳ３４では全ピクセルについて処理を終了としたか判別し、終了していないときは、ステップＳ３１に戻り、全ピクセルの処理をしたら、エンドとする。こうして、輝度によって画像を２値化することにより、炎または煙が存在する可能性がある領域を抽出することができる。
【０１１２】
図２２は他の２値化処理部８の処理を示すフローチャートである。ここでは監視カメラ１が可視カメラであり、特定の色領域を取り出す場合について説明する。
【０１１３】
図２２において、まず、ステップＳ４１では該当ピクセルのＲ成分がＲ成分の上限のしきい値ThR₁以下でかつＲ成分の下限のしきい値ThR₀以上の間にあるか否かを判別し、その範囲内にないときはステップＳ４５に進み、その範囲内にあるときは、ステップＳ４２に進む。
【０１１４】
次に、ステップＳ４２では該当ピクセルのＧ成分がＧ成分の上限のしきい値ThG₁以下でかつＧ成分の下限のしきい値ThG₀以上の間にあるか否かを判別し、その範囲内にないときはステップＳ４５に進み、その範囲内にあるときは、ステップＳ４３に進む。
【０１１５】
次に、ステップＳ４３では該当ピクセルのＢ成分がＢ成分の上限のしきい値ThB₁以下でかつＢ成分の下限のしきい値ThB₀以上の間にあるか否かを判別し、その範囲内にないときはステップＳ４５に進み、その範囲内にあるときは、ステップＳ４４に進む。
【０１１６】
ステップＳ４４では、Ｒ成分、Ｇ成分、Ｂ成分が予め設定された範囲内にある色領域のみを取り出して、該当ピクセルは火災候補領域であると判定する。Ｒ成分、Ｇ成分、Ｂ成分がそれぞれ予め設定された範囲内にないときは、ステップＳ４５で該当ピクセルは火災ではないと判定する。
【０１１７】
次に、ステップＳ４６で全ピクセルについてこの処理が終了したか判別し、終了していないときはステップＳ４１に戻り、終了したら処理をエンドとする。２値化で抽出する色の範囲は、炎を抽出する場合には炎が生成する色範囲を選択し、煙を抽出する場合には、煙によって生じる色範囲を選択する。このように、監視カメラからの入力画像により、色が予め設定された範囲内にある領域のみを取り出し、火災候補領域と判定する。
【０１１８】
図２３は火災判定部９の処理を説明するフローチャートである。
【０１１９】
図２３において、まず、ステップＳ５１で２値化処理部８の結果が火災の条件を満たしているかを判定し、満たしていないときは、ステップＳ５４に進み、満たしているときは、ステップＳ５２に進む。
【０１２０】
次に、ステップＳ５２で該当ピクセルは、ゆらぎ判定部１３でゆらぎと判定されたか否かを判別し、ゆらぎでないと判定されたときは、ステップＳ５４に進み、ゆらぎであると判定されたときは、ステップＳ５３に進む。
【０１２１】
ステップＳ５３では、２値化処理部８で火災の条件を満たしており、ゆらぎであると判定されたので、該当ピクセルは火災候補領域であると判定する。
【０１２２】
ステップＳ５４では、２値化処理部８で火災の条件を満たしておらず、ゆらぎであると判定されていないので、該当ピクセルは火災候補領域でないと判定する。
【０１２３】
次に、ステップＳ５５では全ピクセルについてこの処理が終了したか判別し、終了していないときはステップＳ５１に戻り、終了しているときはステップＳ５６に進む。
【０１２４】
ステップＳ５６では火災候補領域が一定の面積を超えたか否かを判別し、超えていないときは処理を終了とし、超えたときはステップＳ５７に進む。火災候補領域が一定の面積を超えたか否か、例えば、図２４に示すようにフレームメモリ１０上の火災候補領域ＫＬが一定の面積を超えたか否かで判定する。
【０１２５】
ステップＳ５７では火災の発生を判定し、火災通知を行う。警報装置５を作動し、例えばベルを鳴動させ、またモニター装置４上に火災の発生を表示する。
【０１２６】
このように本実施形態においては、安価な装置で、移動物体の影響を受けることなく、炎や煙などのゆらぎをもっている領域を抽出し、精度良く火災を検出することができる。
（付記）
（付記１）
請求項１記載の火災検出装置において、
前記重心位置を計算するための重み係数情報を格納したテーブルを備えたことを特徴とする火災検出装置。
【０１２７】
（付記２）
請求項１又は付記１記載の火災検出装置において、
前記監視カメラからの入力画像の輝度値による物体の高温部のみを抽出する第１の２値化処理手段を備えたことを特徴とする火災検出装置。
【０１２８】
（付記３）
請求項１又は付記１記載の火災検出装置において、
前記監視カメラからの入力画像の色成分により特定の領域のみを抽出する第２の２値化処理手段を備えたことを特徴とする火災検出装置。
【０１２９】
【発明の効果】
以上説明してきたように、本発明によれば、各ピクセル毎の輝度の時間変動と正規分布にモデル化したノイズと比較してピクセルの輝度変動がノイズによって生じたかを判定し、時間軸を円状に配置し、輝度が成す重心位置を求め、ピクセルの輝度変動が物体の移動によって生じたかを判定し、ノイズ判定手段の出力と移動物体判定部の出力に基づいてノイズでなくかつ移動物体でない領域のみをゆらぎ領域として判定するため、移動物体の影響を受けることなく、炎や煙などのゆらぎを持っている領域を抽出し、精度良く火災を検出することができ、信頼性を大幅に向上させることができる。また、安価な装置を利用できるため、コストを大幅に低減することができる。
【０１３０】
また、重心位置を計算するための重み係数情報を格納したテーブルを備えるため、重心位置を計算するとき係数を改めて計算する必要がなく、演算量を低減することができる。
【０１３１】
また、画像を２値化して特定の領域のみを抽出するため、より信頼性の高い火災検出を行うことができる。
【０１３２】
さらに、監視カメラからの入力画像の輝度値により画像を２値化して物体の高温部のみを抽出するため、より信頼性の高い火災検出を行うことができる。
【図面の簡単な説明】
【図１】原理説明図
【図２】本発明の一実施形態を示す全体構成図
【図３】火災検出装置のブロック図
【図４】フレームメモリの記録内容を示す図
【図５】フレームメモリのデータの例を示す図
【図６】輝度の変動を示す図
【図７】ノイズ判定の説明図
【図８】ゆらぎと移動物体の輝度変動を示す図
【図９】時間軸を円形に配置したときの重心位置の説明図
【図１０】重心位置を求める説明図
【図１１】ゆらぎの判定の説明図
【図１２】ウェイトテーブルの格納例を示す図
【図１３】２値化の説明図
【図１４】ノイズ判定部の処理を示すフローチャート
【図１５】有意確率の説明図
【図１６】検定統計量と有意確率の関係を示す図
【図１７】移動物体判定部の処理を示すフローチャート
【図１８】有意確率を示す図
【図１９】検定統計量と有意確率の関係を示す図
【図２０】ゆらぎ判定部の処理を示すフローチャート
【図２１】２値化処理部の処理を示すフローチャート（その一）
【図２２】２値化処理部の処理を示すフローチャート（その二）
【図２３】火災判定部の処理を示すフローチャート
【図２４】火災候補領域の例を示す図
【図２５】従来例の説明図
【図２６】従来のゆらぎ抽出の説明図
【図２７】他の従来例を示すハード構成図
【図２８】図２７のＤＩＰユニットの構成図
【図２９】図２７の処理フローを示す図
【図３０】問題点の説明図
【符号の説明】
１：監視カメラ
２：火災検出装置
３：装置本体
４：モニター装置
５：警報装置
６：ＡＤ変換部
７：ゆらぎ抽出部
８：２値化処理部（第１，第２の２値化処理手段）
９：火災判定部
１０：フレームメモリ（記憶手段）
１１：ノイズ判定部（ノイズ判定手段）
１２：移動物体判定部（移動物体判定手段）
１３：ゆらぎ判定部（ゆらぎ判定手段）
１４：ウェイトテーブル（テーブル）[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a fire detection apparatus that detects a fire by monitoring a wide range outdoors or indoors with an infrared camera or a visible camera and processing the image.
[0002]
There are two methods for detecting fire using images, such as detecting temperature using an infrared camera and extracting fluctuations, but they are inexpensive and can be used to correctly determine a fire without mistaking it as fluctuations. Development of a detection device has been desired.
[0003]
[Prior art]
As a conventional fire detection method using an image, there is a method of detecting temperature using an infrared camera. This uses a sensor capable of measuring temperature with high accuracy to determine a region where the temperature is above a certain level, and uses that as a fire region to detect a fire.
[0004]
For example, as a feature of the heating element captured by the infrared camera, there is no movement of the center of gravity of the pixel, fluctuation of the heating element due to a change in the time difference of the image, and the detection temperature of the image element in advance for each wavelength. The fact that it is within the set predetermined level range is extracted, and the occurrence of fire is determined using the extracted feature as a fire determination criterion.
[0005]
FIG. 25 shows the relationship of wavelength λ-relative intensity l at a certain time at an arbitrary point (area) on the image, and flame spectrum S is represented by a solid curve, and lower limit value α1 for each wavelength band λ1 to λn. When .alpha.n and upper limit values .gamma.1 to .gamma.n are set, it is determined that there is a fire when the shaded portion is within the set level ranges of lower limit values .alpha.1 to .alpha.n and upper limit values .gamma.1 to .gamma.
[0006]
There is also a method of simplifying this and determining that the fire is determined to be within the set level range for the wavelength band of only the peak value.
[0007]
In FIG. 25, when the peak value is the wavelength band λp and the shaded portion is within the set level range of the lower limit value αp and the upper limit value γp, it is determined that a fire has occurred.
[0008]
As another conventional example, there is a method of extracting fluctuations. This is to detect a fire by extracting the inherent fluctuation of a flame using an infrared camera or a visible camera that is inexpensive but not so high in temperature measurement accuracy. In this method, first, an area having a high temperature (in the case of an infrared camera) or an area having a certain range of colors (in the case of a visible camera) is determined, and whether or not the area has fluctuation is determined. If there is, report it as a fire.
[0009]
In the case of an infrared camera, a flame is determined when there is a fluctuation above a certain temperature, and in the case of a visible camera, a flame is determined when there is a fluctuation within a certain color range.
[0010]
Similarly, smoke can be extracted.
[0011]
As a conventional means for extracting the fluctuations of flames and smoke, a method of taking difference images between frames and accumulating the absolute values is used. As shown in FIG. 26, first, a difference is obtained between each frame of the original image, and an image in which the absolute value is obtained is created. Since the fluctuation is a slight change, in order to amplify, the absolute value image of the difference is accumulated to extract the fluctuation.
[0012]
That is, as shown in FIG. 26A, the difference image e between the original image a and the original image b is taken, the difference image f between the original image b and the original image c is taken, and further, the difference image g between the original image c and the original image d is taken. Then, the difference images e, f, and g are accumulated to obtain an accumulated image h. Fluctuations are extracted from the accumulated image h.
[0013]
A hardware configuration example of this method is shown in FIG. In FIG. 27, this conventional example includes a plurality of cameras 101, 102, 103, a plurality of dual-in-line package (DIP) units 104, 105, 106, an image processing (IP) unit 107, and a PSP unit 108. And the host unit 109.
[0014]
A plurality of cameras 101, 102, 103 are switched, images are sequentially input, simple video processing is performed in advance in the DIP units 104, 105, 106, and the processing results of the DIP units 104, 105, 106 are The IP unit 107 performs disturbance removal and feature extraction. Further, it is further processed by the DSP unit 108 dedicated to image processing and sent to the host computer 109.
[0015]
FIG. 28 shows a configuration example of the DIP units 104, 105, and 106. The DIP units 104, 105, and 106 are configured with simple circuits of an AD converter 110, multiplexers 111 to 113, image memories 114 to 118, a subtractor 119, an adder 120, and a latch 121. In order to detect smoke, a CCD camera is connected, and images converted into digital quantities by the AD converter 110 are sequentially stored from the image memory 114 to 117. The subtractor 119 takes the difference between the images and accumulates the obtained difference images using the adder 120 and the image memory 118. FIG. 29 shows a processing flow of smoke detection. First, the DIP units 104, 105, and 106 perform difference image accumulation processing on the image captured from the CCD camera. Next, the result is transferred to the IP unit 107, and disturbance removal and feature extraction are performed collectively.
[0016]
[Problems to be solved by the invention]
However, in such a conventional fire detection method, in the case of a fire detection method using an infrared camera, this method is a highly reliable method, but the sensor is very expensive and expensive. There's a problem.
[0017]
Further, although it is possible to extract fluctuations by using a conventional method for extracting fluctuations, since a change appears in the difference image even when the object moves, there arises a problem that the movement of the object is mistaken as fluctuations. As shown in FIG. 30, when the object moves, a difference image k is obtained from the original image i and the original image j, and the movement of the object is mistaken as fluctuation.
[0018]
The present invention has been made in view of such conventional problems, and is a low-cost fire detection device that accurately detects a fire by extracting fluctuations without being affected by a moving object. The purpose is to provide.
[0019]
[Means for Solving the Problems]
In order to achieve the above object, the present invention is configured as follows (see FIG. 1).
[0020]
In the present invention, when monitoring is performed by the monitoring camera 1 and the image is processed to detect a fire, the time variation of the luminance for each pixel of the input image of the monitoring camera 1 is stored in the storage means 10 to determine noise. The means 11 determines whether or not the luminance variation of the pixel is caused by the noise, by comparing the luminance variation with time for each pixel and the noise modeled in the normal distribution.
[0021]
Further, the moving object determining means 12 arranges the time axis in a circle, obtains the position of the center of gravity formed by the luminance, determines whether the luminance fluctuation of the pixel is caused by the movement of the object, and the fluctuation determining means 13 determines the noise. Based on the output of the means 11 and the output of the moving object determination means 12, only a region that is not noise and is not a moving object is determined as a fluctuation region.
[0022]
In the present invention, a table storing weight coefficient information for calculating the position of the center of gravity is provided.
[0023]
In the present invention, only the high temperature part of the object is extracted by the luminance value of the input image from the monitoring camera 1 by the first binarization processing means.
[0024]
Further, in the present invention, only a specific region is extracted by the color component of the input image from the monitoring camera 1 by the second binarization processing means.
[0025]
According to the present invention having such a configuration, it is determined whether the luminance variation of the pixel is caused by the noise as compared with the temporal variation of the luminance for each pixel and the noise modeled in the normal distribution, and the time axis is set to a circle. The position of the center of gravity formed by the luminance is obtained, it is determined whether the luminance variation of the pixel is caused by the movement of the object, and the movement is not a noise based on the output of the noise determination unit 11 and the output of the moving object determination unit 12 Since only non-object areas are determined as fluctuation areas, areas that have fluctuations such as flames and smoke can be extracted without being affected by moving objects, and fire can be detected with high accuracy, greatly increasing reliability. Can be improved. Further, since an inexpensive device can be used, the cost can be greatly reduced.
[0026]
In addition, since the table storing weight coefficient information for calculating the center of gravity position is provided, it is not necessary to calculate the coefficient again when calculating the center of gravity position, and the amount of calculation can be reduced.
[0027]
In addition, since the image is binarized and only a specific region is extracted, more reliable fire detection can be performed.
[0028]
Furthermore, since the image is binarized based on the luminance value of the input image from the monitoring camera 1 and only the high-temperature part of the object is extracted, more reliable fire detection can be performed.
[0029]
DETAILED DESCRIPTION OF THE INVENTION
FIG. 2 is an overall configuration diagram showing an embodiment of the present invention.
[0030]
In FIG. 2, reference numeral 1 denotes a plurality of monitoring cameras. The monitoring camera 1 monitors a wide range of outdoors and indoors (here, outdoors), and sends an input image to the apparatus body 3 of the fire detection apparatus 2. As the monitoring camera 1, an infrared camera or a visible camera that is inexpensive but does not have a very high temperature measurement accuracy is used. An infrared camera is used to obtain a region having a high temperature, and a visible camera is used to obtain a region having a certain range of color components.
[0031]
The apparatus body 3 performs image processing on the input image from the surveillance camera 1, distinguishes between fluctuation and noise, distinguishes between fluctuation and movement of the object, satisfies the fire condition, determines that there is fluctuation, When is over a certain area, it is determined that a fire has occurred. If it is determined that a fire has occurred, an alarm is displayed on the monitor device 4, and the alarm device 5 notifies the occurrence of the fire with, for example, a bell.
[0032]
FIG. 3 is a block diagram showing the fire detection device 2.
[0033]
In FIG. 3, reference numeral 3 denotes an apparatus main body to which an image from the surveillance camera 1 is input. The apparatus main body 3 is a binary as an AD conversion unit 6, a fluctuation extraction unit 7, and first and second binarization processing means. The processing unit 8 and the fire determination unit 9 are configured.
[0034]
The fluctuation extraction unit 7 includes a frame memory 10 as a storage unit, a noise determination unit 11 as a noise determination unit, a moving object determination unit 12 as a moving object determination unit, a fluctuation determination unit 13 as a fluctuation determination unit, and a weight table (table). ) 14.
[0035]
A plurality of monitoring cameras 1, an alarm device 5, and a monitor device 4 are connected to the apparatus body 3. The monitoring camera 1, the device main body 3, the alarm device 5 and the monitor device 4 constitute a fire detection device 2 as a whole.
[0036]
An input image from the monitoring camera 1 is converted from an analog signal to a digital signal by the AD conversion unit 6, and when the monitoring camera 1 is an infrared camera, the luminance value is sent to the frame memory 10 and the binarization processing unit 8. . When the monitoring camera 1 is a visible camera, color components are sent from the AD conversion unit 6 to the binarization processing unit 8.
[0037]
A frame memory 10 as a storage means stores past input images from the monitoring camera 1. As shown in FIG. 4, the frame memory 10 stores input images at t = t0, t = t1, t = t2,. That is, as shown in FIG. 5, the luminance fluctuation of a certain pixel over a certain time is stored as data.
[0038]
FIG. 6 is a diagram showing the temporal variation of luminance.
[0039]
In FIG. 6, S indicates a certain pixel luminance value on the frame memory 10 and changes on the time axis T.
[0040]
As shown in FIG. 6, paying attention to the temporal change in luminance at each point, it is determined whether or not the fluctuation is based on the change. Specifically, attention is paid to luminance temporal variation at each point on the image over a certain period, and whether or not the point belongs to fluctuation is determined from the characteristic of the variation.
[0041]
When performing the fluctuation determination, it is necessary to make the following two distinctions.
(1) Distinguishing between fluctuation and noise
(2) Distinguishing between fluctuation and object movement
First, the distinction between fluctuation and noise will be described.
[0042]
In order to accurately extract the fluctuation of the flame, it is necessary to remove noise generated from the sensor. Sensor noise can be assumed to be normally distributed. The dispersion value that is the parameter can be measured in advance as a value unique to the sensor. In other words, the nature of noise is known in advance. Therefore, the noise point can be removed by statistically testing whether the sample data observed during a certain period is caused by noise.
[0043]
For this purpose, the distinction between fluctuation and noise is performed as follows.
[0044]
The noise can be assumed to be a normal distribution, and the variance value that is a parameter thereof can be measured in advance as a value unique to the sensor. Therefore, it is statistically tested whether sample data observed during a certain period is caused by noise.
[0045]
Specifically, assuming that the sample data originates from noise, the sum of squares of the sample data is
[0046]
[Expression 1]

[0047]
S / σ ² (Σ ² Is the population = noise variance) with χ of degrees of freedom (N−1) ² According to the distribution (N = number of sample data). Therefore, S / σ ² As the test statistic ² Noise is removed by one-sided testing.
[0048]
That is, as shown in FIG. 7, the test statistic S / σ ² Χ with 29 degrees of freedom ² When following the distribution, the test statistic S / σ ² Is 52 or more, for example, as indicated by an arrow A, it is determined as other than noise, and when it is less than 52, it is determined as noise as indicated by an arrow B.
[0049]
Returning to FIG. 3, as described above, the noise determination unit 11 inputs the time variation of luminance for each pixel recorded in the frame memory 10 and compares it with the noise modeled in the normal distribution. It is determined whether the luminance variation of the pixel is caused by noise.
[0050]
Next, the distinction between fluctuations and objects will be described.
[0051]
In order to stably extract fluctuations due to fire, etc., the influence of the movement of the object must be removed. In order to remove a moving object, it is verified whether data variation is uniform in a certain section. This is because, when the object moves, unlike the fluctuation, nonuniform brightness fluctuation occurs.
[0052]
As shown in FIG. 8A, in the case of fluctuation, the change in the brightness of the flame with respect to the frame is within 130 to 160, and the variation is almost uniform. On the other hand, as shown in FIG. 8B, in the case of a moving object, the luminance change at the time of passing the object with respect to the frame changes greatly from 50 to nearly 200, and the variation is not uniform.
[0053]
Here, each data point is regarded as a weight of uniform weight, and attention is paid to the position of the center of gravity when the section of interest is arranged in a circle. If the variation is uniform, the center of gravity should gather near the center. However, when the luminance variation is caused by a moving object, a non-uniform distribution is generated, so that the center of gravity deviates greatly from the center of the circle. Therefore, it is possible to distinguish the moving object from the fluctuation by measuring how much the position of the center of gravity deviates from the center of the circle.
[0054]
As shown in FIG. 9 (A), in the case of fluctuation, the luminance variation is uniform with respect to the time change, and as shown in FIG. 9 (B), the time axis of the section of interest is shown. When the barycentric position O arranged in a circle and having brightness is obtained, the barycentric position O is near the center of the circle E. On the other hand, as shown in FIG. 9C, when the object is a moving object, the luminance with respect to the time change greatly changes, and the variation is large. As shown in FIG. 9D, when the time axis of the section of interest is arranged in a circle and the barycentric position O at which brightness is obtained is obtained, the barycentric position O deviates greatly from the center of the circle E.
[0055]
As shown in FIG. 10A, the case where the data of the luminance r1 to r4 has four data D1 to D4 will be described.
[0056]
As shown in FIG. 10B, the time axis is arranged in a circle, and the data D1 of the luminance r1 perpendicular to the center M of the circle E, the data D2 of the luminance r2 on the right side of 90 degrees, the data D2 of the luminance r2 below 180 degrees If the data D3 of the luminance r3 and the data D4 of the luminance r4 are arranged on the left side by 90 degrees, the gravity center position O is slightly below the center M of the circle E.
[0057]
As described above, when the measured data fluctuates, unlike the moving object, the data variation in the fixed section is uniform, and thus the uniformity is verified.
[0058]
The data section of interest is arranged in a circle, each data point is regarded as a weight, and whether or not the fluctuation is determined by the position of the center of gravity formed by the weight.
[0059]
The barycentric coordinates when the data are arranged in a circle (X _m , Y _m ). Then
[0060]
[Expression 2]

[0061]
It becomes. Where V ₀ , V ₁ ... Represents the luminance at each point. N is the number of data used for evaluation.
[0062]
Assume that the point you are thinking of belongs to "fluctuation". Here, each brightness in fluctuation, V ₀ , V ₁ ... is a normal distribution, Nd (μ, σ ² ).
[0063]
Then X _m It follows that follows the normal distribution
[0064]
[Equation 3]

[0065]
Therefore, by converting as follows, X _m Canonicalized into a normal distribution with mean = 0 and standard deviation = 1.
[0066]
[Expression 4]

[0067]
Similarly Y _m Y converted _m When the sum of squares R is obtained, R is χ of 2 degrees of freedom. ² Follow the distribution.
[0068]
[Equation 5]

[0069]
Whether or not the fluctuation is present is determined by testing the assumption that “the point being considered is fluctuation” by using R as a test statistic and the significance level α.
[0070]
FIG. 11 is a graph showing the relationship between the fluctuation test statistic R and the probability density.
[0071]
In FIG. 11, the test statistic R is χ with 2 degrees of freedom. ² Follow the distribution. In this example, when the test statistic R is 9 or more, for example, it is determined as a moving object as indicated by an arrow G, and when it is less than 9, it is determined as fluctuation as indicated by an arrow H.
[0072]
Returning to FIG. 3, as described above, the moving object determination unit 12 arranges the time axis in a circle and obtains the center of gravity position where the luminance is formed, and the luminance variation of the pixel is caused by the movement of the object by the center of gravity position. Judge whether it is due to fluctuation or fluctuation.
[0073]
The moving object determination unit 12 includes a weight table (table) 14, and the center of gravity position (X _m , Y _m ) Is calculated in advance for the coefficient portion and stored in the weight table 14. As shown in FIG. 12A, the weight table 14 includes data cos (2π / N × 0), data cos (2π / N × 1), data cos (2π / N × 2) · .. Are stored, and as shown in FIG. 12B, the weight table 14 has data sin (2π / N × 0), data sin (2π / N × 1), data sin (2π /) for the Y coordinate. N × 2)... Are stored.
[0074]
Thus, in (Formula 1), (X _m , Y _m ) Can be determined in advance, the coefficient part is calculated in advance and recorded in the weight table 14. (X _m , Y _m ), The calculation amount can be reduced by referring to the values from the table 14 instead of calculating the coefficients anew.
[0075]
The output of the moving object determination unit 12 is sent to the fluctuation determination unit 13, and the output of the noise determination unit 11 is also sent to the fluctuation determination unit 13. The fluctuation determination unit 13 integrates the output of the noise determination unit 11 and the output of the moving object determination unit 12 and extracts only a region that is not noise and is not a moving object as a fluctuation region.
[0076]
In the case of an infrared camera, the binarization processing unit 8 inputs data obtained by converting an input image into a luminance value by the AD conversion unit 6 from the monitoring camera 1 and extracts only a high temperature part of the object.
[0077]
That is, the binarization processing unit 8 binarizes the image by extracting an area having a luminance value equal to or higher than a certain value from the input image from the monitoring camera 1, and binarizes the area and the fluctuation extraction unit. The common part of the fluctuation area detected in step 1 is extracted as a fire area.
[0078]
As the range of luminance extracted by binarization, the range of luminance generated by flame is selected when detecting a flame, and the range of luminance generated by smoke is selected when extracting smoke.
[0079]
As shown in FIG. 13A, when there is a luminance signal as shown by S, a region X exceeding the threshold value T is taken out and is set as a binarized signal Y as shown in FIG.
[0080]
Further, in the case of a visible camera, the binarization processing unit 8 extracts only a specific area from the color component of the input image from the monitoring camera 1.
[0081]
In other words, the binarization processing unit 8 binarizes the image by taking out only the region in which the color is within a preset range from the input image from the surveillance camera 1, and fluctuates the binarized region and fluctuation. A common part of the fluctuation areas detected by the extraction unit is extracted as a fire area.
[0082]
As the range of colors extracted by binarization, the color range generated by the flame is selected when extracting the flame, and the color range generated by the smoke is selected when extracting the smoke.
[0083]
This binarization process is also performed by a method as shown in FIG.
[0084]
The output of the binarization processing unit 8 is sent to the fire determination unit 9, and the output of the fluctuation determination unit 13 is also sent to the fire determination unit 9. The fire determination unit 9 performs final determination on the results output from the fluctuation determination unit 13 and the binarization processing unit 8. First, it is confirmed that the result of the binarization processing unit 8 satisfies the fire condition, and the pixel determined to be fluctuation by the fluctuation determination unit 13 is set as a fire candidate area. When this fire candidate area exceeds a certain area, it is determined that a fire has occurred and an alarm is issued. The alarm device 5 rings a bell to notify the occurrence of a fire, and the monitor device 4 displays the occurrence of a fire.
[0085]
FIG. 14 is a flowchart showing processing of the noise determination unit 11.
[0086]
In FIG. 14, first, in step S1, the sum of squares S is calculated for each pixel from the data (luminance value) in the frame memory 10.
[0087]
[Formula 6]

[0088]
Next, in step S2, the test statistic S / σ ² Χ with degrees of freedom (N-1) ² The significance P is obtained from the distribution. σ ² Is the variance of noise, and N is the number of sample data.
[0089]
As shown in FIG. 15, the area (= probability) indicated by ME indicates the significance probability P. For example, as indicated by the arrow Z, the currently measured test statistic S / σ ² When the value of is, for example, 35, the significance probability P is 20%. FIG. 16 shows the test statistic S / σ ² And the significance probability P. For example, the test statistic S / σ ² 1, 5, 10, the significance probability P is 100%, 20 is 89%, 30 is 41%, 40 is 8%, and 50 is 1%.
[0090]
Next, the significance probability P calculated in step S3 is compared with a preset significance level α.
[0091]
Next, if the significance probability P is smaller than the significance level α, it is determined in step S4 that the corresponding pixel is not noise. If the significance probability P is greater than the significance level α, it is determined in step S5 that the corresponding pixel is noise. For example, when the significance level α is 3%, it is determined that the noise is not noise when the significance probability P is 3% or less, and is determined to be noise when the significance probability P exceeds 3%.
[0092]
Next, when the process has not been completed for all pixels in step S6, the process returns to step S1, and when the process is completed, the process is ended.
[0093]
FIG. 17 is a flowchart showing the processing of the moving object determination unit 12.
[0094]
In FIG. 17, first, in step S11, the center of gravity (X _m , Y _m ) Center of gravity (X _m , Y _m ) Is obtained from Equation 1. At this time, the coefficient portion (see FIG. 12) stored in advance in the weight table 14 is extracted and calculated to reduce the amount of calculation.
[0095]
Next, in step S12, the test statistic R is χ with 2 degrees of freedom. ² The significance probability P is obtained from following the distribution. The test statistic R is obtained from Equation 4. Equation 4 is
[0096]
[Expression 7]

[0097]
It is.
[0098]
The test statistic R is χ with 2 degrees of freedom as shown in FIG. ² Follow the distribution. In FIG. 18, the area (= probability) formed by the region indicated by ME indicates the significance probability P. As indicated by an arrow Z, if the currently measured value of the test statistic R is 4, for example, the significance probability P is 14%.
[0099]
FIG. 19 shows the relationship between the test statistic R and the significance probability P. For example, when the test statistic R is 1, the significance probability P is 61%, 3 is 22%, 5 is 8%, 7 is 3%, and 9 is 18%.
[0100]
Next, in step S13, the calculated significance probability P is compared with a preset significance level α. If the significance probability P is smaller than the significance level α, it is determined in step S14 that the pixel is a moving object. If the significance probability P is greater than the significance level α, it is determined in step S15 that the pixel is not a moving object. . For example, when the significance level α is set to 2%, when the significance probability P is 2% or less, it is determined that the corresponding pixel is a moving object, and when the significance probability exceeds 2%, the corresponding pixel is not a moving object. judge.
[0101]
Next, in step S16, it is determined whether or not this process has been completed for all pixels. If it has not been completed, the process returns to step S11.
[0102]
FIG. 20 is a flowchart showing the processing of the fluctuation determination unit 13.
[0103]
In FIG. 20, first, in step S21, it is determined from the output of the noise determination unit 11 whether or not the corresponding pixel is noise. If it is noise, the process proceeds to step S24. If not, the process proceeds to step S22.
[0104]
In step S22, it is determined from the output of the moving object determination unit 12 whether the corresponding pixel is a moving object. If it is a moving object, the process proceeds to step S24. If it is not a moving object, the process proceeds to step S23. In step S23, since the corresponding pixel is not noise and is not a moving object, it is determined that the corresponding pixel is fluctuating.
[0105]
In step S24, since the corresponding pixel is a noise or a moving object, it is determined that the corresponding pixel is not a fluctuation.
[0106]
In step S25, it is determined whether or not the processing for all the pixels has been completed. If the processing has not been completed, the processing returns to step S21. In this way, only a region that is neither noise nor a moving object is extracted as a fluctuation region.
[0107]
FIG. 21 is a flowchart showing the processing of the binarization processing unit 8. Here, the case where the surveillance camera 1 is an infrared camera and a high temperature part is taken out will be described.
[0108]
In FIG. 21, first, in step S31, it is determined whether or not the luminance value of the corresponding pixel is equal to or higher than a certain luminance threshold value. When the luminance value is less than the certain threshold value, the process proceeds to step S33. If it is greater than or equal to the threshold value, the process proceeds to step S32.
[0109]
In step S32, it is determined that the corresponding pixel is a fire candidate area. That is, by extracting an area having a luminance value equal to or higher than a certain value from the input image from the monitoring camera 1, the image is binarized, and the binarized area is set as a fire candidate area.
[0110]
As the range of luminance extracted by binarization, the range of luminance generated by the flame is selected when detecting a flame, and the range of luminance generated by smoke is selected when extracting smoke. When the luminance value of the pixel is less than a certain threshold value, it is determined that the pixel is not a fire.
[0111]
In step S34, it is determined whether or not processing has been completed for all pixels. If not, processing returns to step S31, and processing is completed for all pixels. Thus, by binarizing the image based on the luminance, it is possible to extract a region where flame or smoke may exist.
[0112]
FIG. 22 is a flowchart showing processing of another binarization processing unit 8. Here, a case where the monitoring camera 1 is a visible camera and a specific color area is extracted will be described.
[0113]
In FIG. 22, first, in step S41, the R component of the corresponding pixel is the upper threshold value ThR of the R component. ₁ Threshold ThR below the lower limit of the R component ₀ It is determined whether or not it is within the above range. If it is not within the range, the process proceeds to step S45. If it is within the range, the process proceeds to step S42.
[0114]
Next, in step S42, the G component of the corresponding pixel is the upper limit threshold ThG of the G component. ₁ The lower threshold ThG of the G component below ₀ It is determined whether or not it is within the above range. If it is not within the range, the process proceeds to step S45. If it is within the range, the process proceeds to step S43.
[0115]
Next, in step S43, the B component of the corresponding pixel is the upper threshold value ThB of the B component. ₁ Threshold ThB below the lower limit of the B component ₀ It is determined whether or not it is within the above range. If it is not within the range, the process proceeds to step S45. If it is within the range, the process proceeds to step S44.
[0116]
In step S44, only the color region in which the R component, G component, and B component are within a preset range is extracted, and it is determined that the corresponding pixel is a fire candidate region. If the R component, G component, and B component are not within the preset ranges, it is determined in step S45 that the pixel is not a fire.
[0117]
Next, in step S46, it is determined whether or not this process has been completed for all the pixels. If not, the process returns to step S41. As the color range extracted by binarization, the color range generated by the flame is selected when extracting the flame, and the color range generated by the smoke is selected when extracting the smoke. In this way, only an area whose color is within a preset range is extracted from the input image from the surveillance camera, and is determined as a fire candidate area.
[0118]
FIG. 23 is a flowchart for explaining the processing of the fire determination unit 9.
[0119]
In FIG. 23, first, in step S51, it is determined whether or not the result of the binarization processing unit 8 satisfies the fire condition. If not, the process proceeds to step S54. If the result is satisfied, the process proceeds to step S52. .
[0120]
Next, in step S52, it is determined whether or not the pixel is determined to be fluctuation by the fluctuation determination unit 13. When it is determined that the pixel is not fluctuation, the process proceeds to step S54. Proceed to step S53.
[0121]
In step S53, since the binarization processing unit 8 satisfies the fire condition and is determined to be fluctuation, it is determined that the corresponding pixel is a fire candidate area.
[0122]
In step S54, since the binarization processing unit 8 does not satisfy the fire condition and is not determined to be a fluctuation, the pixel is determined not to be a fire candidate area.
[0123]
Next, in step S55, it is determined whether or not this process has been completed for all pixels. If not, the process returns to step S51. If completed, the process proceeds to step S56.
[0124]
In step S56, it is determined whether or not the fire candidate area exceeds a certain area. If it does not exceed, the process ends. If it exceeds, the process proceeds to step S57. It is determined whether or not the candidate fire area exceeds a certain area, for example, whether or not the candidate fire area KL on the frame memory 10 exceeds a certain area as shown in FIG.
[0125]
In step S57, the occurrence of a fire is determined and a fire notification is given. The alarm device 5 is activated, for example, a bell is sounded, and the occurrence of a fire is displayed on the monitor device 4.
[0126]
As described above, in this embodiment, an inexpensive device can extract a region having fluctuations such as flame and smoke without being affected by a moving object, and can detect a fire with high accuracy.
(Appendix)
(Appendix 1)
The fire detection device according to claim 1,
A fire detection apparatus comprising a table storing weight coefficient information for calculating the position of the center of gravity.
[0127]
(Appendix 2)
In the fire detection device according to claim 1 or appendix 1,
A fire detection apparatus comprising first binarization processing means for extracting only a high temperature portion of an object based on a luminance value of an input image from the monitoring camera.
[0128]
(Appendix 3)
In the fire detection device according to claim 1 or appendix 1,
A fire detection apparatus comprising second binarization processing means for extracting only a specific area based on a color component of an input image from the monitoring camera.
[0129]
【The invention's effect】
As described above, according to the present invention, it is determined whether the luminance variation of a pixel is caused by noise compared with the temporal variation of luminance for each pixel and noise modeled in a normal distribution, and the time axis is set to a circle. The center of gravity position is determined, and it is determined whether the luminance fluctuation of the pixel is caused by the movement of the object. Based on the output of the noise determination means and the output of the moving object determination unit, it is not noise and is not a moving object Since only the area is determined as a fluctuation area, it is possible to extract areas that have fluctuations such as flames and smoke without being affected by moving objects, and to detect fires with high accuracy, greatly improving reliability. Can be made. Further, since an inexpensive device can be used, the cost can be greatly reduced.
[0130]
In addition, since the table storing weight coefficient information for calculating the center of gravity position is provided, it is not necessary to calculate the coefficient again when calculating the center of gravity position, and the amount of calculation can be reduced.
[0131]
In addition, since the image is binarized and only a specific region is extracted, more reliable fire detection can be performed.
[0132]
Furthermore, since the image is binarized based on the luminance value of the input image from the surveillance camera and only the high temperature portion of the object is extracted, more reliable fire detection can be performed.
[Brief description of the drawings]
FIG. 1 illustrates the principle
FIG. 2 is an overall configuration diagram showing an embodiment of the present invention.
[Figure 3] Block diagram of the fire detection device
FIG. 4 is a diagram showing recorded contents of a frame memory
FIG. 5 is a diagram showing an example of data in a frame memory
FIG. 6 is a diagram showing variations in luminance.
FIG. 7 is an explanatory diagram of noise determination.
FIG. 8 is a diagram showing fluctuation and luminance fluctuation of a moving object.
FIG. 9 is an explanatory diagram of the position of the center of gravity when the time axis is arranged in a circle.
FIG. 10 is an explanatory diagram for obtaining the position of the center of gravity.
FIG. 11 is an explanatory diagram of fluctuation determination
FIG. 12 is a diagram showing an example of storing a weight table
FIG. 13 is an explanatory diagram of binarization.
FIG. 14 is a flowchart illustrating processing of a noise determination unit.
FIG. 15 is an explanatory diagram of significance probability.
FIG. 16 is a diagram showing the relationship between the test statistic and the significance probability
FIG. 17 is a flowchart illustrating processing of a moving object determination unit.
FIG. 18 is a diagram showing significance.
FIG. 19 is a graph showing the relationship between test statistics and significance probability.
FIG. 20 is a flowchart illustrating processing of a fluctuation determination unit.
FIG. 21 is a flowchart (part 1) showing processing of a binarization processing unit.
FIG. 22 is a flowchart showing the processing of the binarization processing unit (part 2).
FIG. 23 is a flowchart showing processing of a fire determination unit
FIG. 24 is a diagram showing an example of a fire candidate area
FIG. 25 is an explanatory diagram of a conventional example.
FIG. 26 is an explanatory diagram of conventional fluctuation extraction.
FIG. 27 is a hardware configuration diagram showing another conventional example.
28 is a block diagram of the DIP unit in FIG. 27.
FIG. 29 is a diagram showing a processing flow of FIG. 27;
FIG. 30 is an explanatory diagram of problems.
[Explanation of symbols]
1: Surveillance camera
2: Fire detection device
3: Main unit
4: Monitor device
5: Alarm device
6: AD converter
7: Fluctuation extraction unit
8: Binarization processing unit (first and second binarization processing means)
9: Fire judgment part
10: Frame memory (storage means)
11: Noise determination unit (noise determination means)
12: Moving object determination unit (moving object determination means)
13: Fluctuation determination unit (fluctuation determination means)
14: Weight table (table)

Claims (1)

In a fire detection device that detects fire by monitoring with a surveillance camera and processing the image,
Storage means for storing temporal variation of luminance for each pixel of the input image of the surveillance camera;
Noise determining means for determining whether the luminance variation of the pixel is caused by noise in comparison with the time variation of luminance for each pixel and noise modeled in a normal distribution;
The time axis is arranged in a circle, the barycentric position formed by the luminance is obtained, the test statistic is obtained from the barycentric position, the significance is obtained from the test statistic, and the significance is smaller than a preset significance level A moving object determination means for determining whether the luminance variation of the pixel is caused by the movement of the object,
Fluctuation determination means for determining only a region that is not noise and is not a moving object based on the output of the noise determination unit and the output of the moving object determination unit;
A binarization processing unit that binarizes an area in which a luminance value of an input image of the camera is equal to or greater than a certain value;
A fire has occurred when the part determined to be fluctuation by the fluctuation determination unit and the part where the luminance value is equal to or greater than a certain value in the binarization processing unit are designated as fire candidate areas, and the fire candidate area is equal to or greater than a certain area. A fire determination unit that issues an alarm by judging
A fire detection device comprising:

Images