JP7764740B2 - Radar tracking method, noise removal method, device and equipment - Google Patents

Description

The present invention relates to the field of information technology, and in particular to radar tracking methods, noise reduction methods, devices, and equipment.

Radar can sense (detect) information such as the distance, speed, and azimuth of moving objects through radio signals and form a point cloud. Conventional radar tracking algorithms can track objects by performing operations such as clustering and filtering based on the spatial information of the point cloud. However, at long distances (when the distance between the moving object and the radar is large), the accuracy of the radar's spatial measurements deteriorates, the point cloud tends to disperse, and clustering becomes impossible.

To solve the above-mentioned problems, embodiments of the present invention aim to provide a radar tracking method, a noise removal method, a device, and an apparatus.

According to a first aspect of an embodiment of the present invention, there is provided a radar tracking method, the method comprising:
Perform clustering on the point cloud acquired by radar detection in the range-Doppler plane to obtain a range-Doppler point cloud cluster;
performing secondary clustering on the distance-Doppler point cloud clusters in a spatial dimension or an angular dimension to obtain secondary clustered point cloud clusters;
and determining a tracked object based on the range-Doppler point cloud cluster and its corresponding secondary clustered point cloud cluster.

According to a second aspect of an embodiment of the present invention, there is provided a method for denoising, the method comprising:
Perform statistics on point cloud noise information on the range-Doppler plane, where the point cloud noise information includes a noise statistical total number n _n and a number n _r,v of noise points appearing at each position on the range-Doppler plane;
Analyzing the point cloud clusters on the distance-Doppler plane based on the point cloud noise information to determine whether the point cloud clusters are point cloud clusters caused by noise; and if the point cloud clusters are point cloud clusters caused by noise, removing the point cloud clusters on the distance-Doppler plane caused by noise.

According to a third aspect of an embodiment of the present invention, there is provided a radar tracking apparatus, the apparatus comprising:
a first clustering unit that performs clustering on the point cloud acquired by radar detection in a range-Doppler plane to obtain a range-Doppler point cloud cluster;
a second clustering unit for performing secondary clustering on the range-Doppler point cloud clusters in a spatial dimension or an angular dimension to obtain secondary clustered point cloud clusters; and a determination unit for determining a tracked object based on the range-Doppler point cloud clusters and their corresponding secondary clustered point cloud clusters.

According to a fourth aspect of an embodiment of the present invention, there is provided a noise removal apparatus, the apparatus comprising:
a statistical unit for performing statistics on point cloud noise information in a range-Doppler plane, the point cloud noise information including: a noise statistical total number n _n and a number n _r,v of noise points appearing at each position in the range-Doppler plane;
An analysis unit that analyzes the point cloud clusters on the distance-Doppler plane based on the point cloud noise information and determines whether the point cloud clusters are point cloud clusters caused by noise; and a processing unit that, when the point cloud clusters are point cloud clusters caused by noise, removes the point cloud clusters on the distance-Doppler plane caused by noise.

According to another aspect of an embodiment of the present invention, there is provided a computing device, the computing device including a processor and a memory, the memory storing a computer program, and the processor configured to execute the computer program to implement the method described in the first or second aspect.

The advantageous effects of the present invention are as follows: on the one hand, the double clustering strategy improves tracking accuracy and extends the effective tracking range, solving the problem of poor radar spatial measurement accuracy at long distances, point clouds easily becoming dispersed, and clustering not being possible; on the other hand, the accuracy of spatial measurement can be further improved by removing point cloud clusters caused by noise on the range-Doppler plane.

FIG. 2 illustrates an example of a radar tracking method according to an embodiment of the present invention. One indication of the range-Doppler plane; FIG. 10 is a diagram illustrating an example of removing noise point cloud clusters in a range-Doppler plane. FIG. 10 is a diagram illustrating an example of updating point cloud noise information on a range-Doppler plane. FIG. 1 shows a radar point cloud in the X-Y plane. FIG. 1 illustrates a radar point cloud in the range-Doppler plane. FIG. 10 illustrates an example of performing secondary clustering on range-Doppler point cloud clusters in the spatial dimension. FIG. 10 is a diagram illustrating an example of correcting a spatial position. FIG. 10 illustrates an example of performing secondary clustering on range-Doppler point cloud clusters in the angle dimension. FIG. 10 illustrates an example of determining a tracking object. FIG. 10 is a diagram illustrating an example of a noise removal method according to an embodiment of the present invention. 1 is a diagram illustrating an example of a radar tracking device according to an embodiment of the present invention. 1 is a diagram illustrating an example of a noise removal device according to an embodiment of the present invention. FIG. 1 illustrates a computer device according to an embodiment of the present invention.

Below, preferred embodiments for carrying out the present invention will be described in detail with reference to the accompanying drawings. Note that these embodiments are merely illustrative and do not limit the present invention.

<Example of the first aspect>
An embodiment of the present invention provides a radar tracking method, which is shown in Figure 1 as an example of the radar tracking method in an embodiment of the present invention. As shown in Figure 1, the method includes the following steps:

101: Perform clustering on the point cloud acquired by radar detection in the range-Doppler plane to obtain a range-Doppler point cloud cluster;
102: Perform secondary clustering on the range-Doppler point cloud cluster in a spatial dimension or an angular dimension to obtain a secondary clustered point cloud cluster;
103: Determine a tracking object based on the range-Doppler point cloud cluster and its corresponding secondary clustered point cloud cluster.

The method in this embodiment of the present invention improves tracking accuracy and extends the effective tracking range through a two-step clustering strategy, solving the problem of poor radar spatial measurement accuracy at long distances, which makes point clouds prone to dispersion and makes clustering impossible.

In an embodiment of the present invention, a radar periodically transmits a radio signal into space, receives signals reflected by objects in space (referred to as reflected signals), and analyzes the reflected signals to output a point cloud, which includes spatial position and velocity information of objects within the radar's coverage. The point cloud output by the radar (referred to as radar point cloud) is represented as {o _i , 1≦i≦n _o }, where o _i is the ith point and n _o is the number of points in the point cloud.

Wherein, point o _i in the radar point cloud can be represented as o _i = (r _i , v _i , x _i , y _i , z _i , α _i , β _i ), where r _i is the distance frequency point of point o _i , v _i is the Doppler velocity frequency point of point o _i , (x _i , y _i , z _i ) is the spatial coordinate of point o _i , and α _i and β _i are the horizontal and vertical angular frequency points of point o _i, respectively. Wherein, distance frequency point r _i is used to calculate the distance from point o _i to the radar, and when the distance resolution of the radar is Δ _r , the distance from point o _i to the radar can be expressed as r _i Δ _r . Wherein, Doppler velocity frequency point v _i is used to calculate the velocity of point o _i relative to the radar, and when the velocity resolution of the radar is Δ _v , the Doppler velocity of point o _i relative to the radar is v _i Δ _v . The maximum number of distance frequency points and the maximum number of Doppler velocity frequency points of the radar are denoted by n _r and n _v, respectively. The horizontal angle frequency point _αi is used to calculate the horizontal angle _A of point _oi relative to the radar, A= _waαi /n _a , where n _a is the total number of frequency points of the horizontal angle FFT, and w _a is a parameter related to the radar's multi-antenna configuration. Similarly to the horizontal angle frequency point _αi , the vertical angle frequency point _βi can be used to calculate the vertical angle of point _oi relative to the radar.

In some implementations of the present invention, in 101, clustering the point cloud in the distance-Doppler plane involves mapping the point cloud onto the distance-Doppler plane, and dividing the point cloud into a plurality of point cloud clusters (referred to as distance-Doppler point cloud clusters) and non-clustered point clouds (points that cannot form a point cloud cluster are referred to as "non-clustered point clouds") based on the distance of the point cloud in the distance-Doppler plane, where the number of points in each point cloud cluster is greater than a first threshold _Tn1 , and the distance from each point in each point cloud cluster to the point cloud cluster is less than a second threshold _Td1 .

In the above-described embodiment, the range-Doppler plane is a two-dimensional plane whose horizontal axis represents distance frequency points and whose vertical axis represents Doppler velocity frequency points. The radar point cloud can be mapped onto the range-Doppler plane based on the distance frequency points and Doppler velocity frequency points of each point in the radar point cloud.

Figure 2 shows an example of a radar point cloud mapped onto a range-Doppler plane. The points in area A are numerous and concentrated, which may be due to moving objects. The points in area B are few and dispersed, which may be due to noise.

In the above embodiment, the clustering operation is performed in the distance-Doppler plane, and the point cloud is divided into a number of clusters (referred to as distance-Doppler point cloud clusters) based on the distance of the point cloud in the distance-Doppler plane. After distance-Doppler clustering is performed, all points in the point cloud can be divided into a number of distance-Doppler point cloud clusters and non-clustered points. Among them, the points in the non-clustered point cloud may be due to noise.

In the above embodiment, a range-Doppler point cloud cluster consists of multiple points that are close to each other on the range-Doppler plane, among which the number of points is greater than a threshold T _n1 (referred to as the first threshold) and the distance from each point to the point cloud cluster is less than a threshold T _d1 (referred to as the second threshold).

For example, let C={o _k ,1≦k≦nc} represent a distance-Doppler point cloud cluster, where n _c is the number of points in the point cloud cluster, o _k is the kth point belonging to the point cloud cluster, in which n _c >T _n1 , and the distance d(o _k ,C) from point o _k to point cloud cluster C is less than a threshold T _d1 , i.e., d(o _k ,C)<T _d1 .

In some embodiments, the distance from each point in a point cloud cluster to that point cloud cluster refers to the minimum distance from that point to all other points in that point cloud cluster.

For example, the distance d(o _k , C) from point o _k to point cloud cluster C is the minimum value of the distance between point o _k and other points in point cloud cluster C,
In this expression, d(o _j ,o _k ) refers to the distance between o _j and o _k —the distance in the Doppler plane.

In some embodiments, the distance between two points in a point cloud cluster is a weighted sum of the absolute value of the difference in the distance frequency points of the two points and the absolute value of the difference in the Doppler velocity frequency points of the two points.

For example, the distance between points o _j and o _k in the point cloud cluster C—the distance in the Doppler plane, d(o _j , o _k )—is the weighted sum of the absolute value of the distance frequency point difference and the absolute value of the Doppler velocity frequency point difference between points o _j and o _k , i.e.,
d(o _j ,o _k )=a|r _j -r _k |+(1-a)|v _j -v _k | (1)
is.

Wherein, r _j and r _k are the distance frequency points of the point groups o _j and o _k , respectively, v _j and v _k are the Doppler velocity frequency points of the point groups o _j and o _k , respectively, and a is a predetermined weight value (weight) ranging from 0 to 1.

Note that the above formula (1) for calculating the distance between points within a point cloud cluster is merely an example, and the present invention is not limited to this. The distance between points within a point cloud cluster may also be calculated using other conventional distance calculation methods, such as a Euclidean distance calculation algorithm.

In the embodiment of the present invention, the specific algorithm for performing clustering on the distance-Doppler plane is not limited, and for example, DBSCAN (Density-Based Spatial Clustering of
Conventional clustering methods such as IEEE Standards for Applications with Noise (GSOM) and OPTICS (Ordering Points To Identify the Clustering Structure) may also be employed.

In some embodiments, as shown in FIG. 1, the method may further include the following steps:

104: Remove point cloud clusters due to noise from the range-Doppler point cloud clusters.

In some embodiments, as shown in FIG. 1, 104 is performed after 101, i.e., after obtaining the range-Doppler point cloud clusters by 101, the noisy point cloud clusters in the range-Doppler point cloud clusters can be removed. Removing the noisy point cloud clusters can further improve the accuracy of the spatial measurement of radar tracking.

In some embodiments, after 102, an operation for removing noise-caused point cloud clusters within the secondary clustered point cloud clusters may be added. The specific operation flow is similar to that of 104, so hereinafter, only 104 will be used as an example to describe the method for removing noise-caused point cloud clusters within the point cloud clusters.

Figure 3 shows an example of removing point cloud clusters due to noise in range-Doppler point cloud clusters. As shown in Figure 3, the method includes the following steps:

301: Perform statistics on point cloud noise information on the range-Doppler plane, the point cloud noise information including: a total noise statistical count n _n and a number of times n _r,v that a noise point appears at each position on the range-Doppler plane;
302: Analyzing the range-Doppler point cloud cluster based on the point cloud noise information to determine whether the point cloud cluster is a noise-induced point cloud cluster;
303: If the point cloud cluster is a noisy point cloud cluster, remove the noisy point cloud cluster in the range-Doppler plane.

In some embodiments of the present invention, the point cloud noise information on the range-Doppler plane includes the total number of noise statistics and the number of times that noise points appear at each position on the plane. In 302, in some implementations, determining whether a range-Doppler point cloud cluster (e.g., point cloud cluster C) is a noise point cloud cluster (referred to as a noise point cloud cluster) can be performed as follows: based on the above point cloud noise information, calculate the probability (represented as P(C)) that the point cloud cluster C is noise; if the probability P(C) that the point cloud cluster C is noise is greater than a threshold _Tn (referred to as a third threshold), determine that the point cloud cluster C is a noise point cloud cluster; otherwise, determine that the point cloud cluster C is a moving object.

In some embodiments, the probability that a range-Doppler point cloud cluster is a noise point cloud cluster is the average probability that all points belonging to the point cloud cluster are noise. For example, still taking point cloud cluster C as an example, the probability P(C) that point cloud cluster C is noise is the average probability that each point in point cloud cluster C is noise:
It is expressed as:

Wherein, C is the range-Doppler point cloud cluster, o _k is a point in the point cloud cluster C, n _C is the number of points in the point cloud cluster C; P(C) is the probability that the range-Doppler point cloud cluster C is a noise point cloud cluster; P(o _k ) is the probability that point o _k is noise.

In some embodiments, the probability that each point in point cloud cluster C is noise is the ratio of the number of times that a noise point appears at each location in the range-Doppler plane to the total number of times the noise statistic appears.

For example, if the total number of noise statistics is represented by n _n , the number of times a noise point appears at each position (the position of distance frequency point r and Doppler frequency point v) on the range-Doppler plane is represented by n _r,v , and the probability of a noise point appearing at that position is represented by P(r,v), then:
P(r,v)=n _r,v /n _n (3)
is.

In equation (2), P(o _k )=P(r _k ,v _k ), where r _k and v _k are the distance frequency point and the Doppler velocity frequency point of point o _k , respectively.

In some embodiments of the present invention, as shown in FIG. 3, the method further includes the following steps:

304: Update point cloud noise information in the range-Doppler plane, which includes adding 1 to the noise statistics total count and adding 1 to the count of noise occurrences at positions in the range-Doppler plane corresponding to all points belonging to the range-Doppler noise point cloud cluster and all points in the non-cluster point cloud.

Figure 4 shows an example of updating point cloud noise information in the range-Doppler plane. As shown in Figure 4, the method includes the following steps:

401: Update the total number of noise statistics and add 1 to the total number of noise statistics;
402-405: Update the number of times noise appears at each position point on the range-Doppler plane.

In the example of Figure 4, (r, v) represents a point at one position on the distance-Doppler plane, where r is the distance frequency point and v is the Doppler velocity frequency point. In 403, it is determined whether a point exists in the distance-Doppler noise point cloud cluster or non-cluster point cloud where the distance frequency point is r and the Doppler velocity frequency point is v. If so, the operation of 404 is performed; if not, no processing is performed. In 404, the number of times noise appears at the position (r, v) is updated, and the number of times noise appears at that position is added by 1.

Note that while the above explanation has focused on the removal of noise-caused point cloud clusters from the range-Doppler point cloud cluster obtained in 101, a similar method can also be applied to the removal of noise-caused point cloud clusters from the secondary clustered point cloud cluster obtained in 102. A detailed explanation of this will be omitted here.

Furthermore, while only one method for removing point cloud clusters caused by noise has been described above, the present invention is not limited to this. In embodiments of the present invention, point cloud clusters caused by noise can also be removed using other feasible methods. Detailed descriptions of these methods will be omitted here.

In an embodiment of the present invention, in 102, secondary clustering may be performed on the above-mentioned range-Doppler point cloud clusters in the spatial dimension, or secondary clustering may be performed on the above-mentioned range-Doppler point cloud clusters in the angular dimension, to obtain secondary clustered point cloud clusters.

In an embodiment of the present invention, if two moving objects have the same distance and velocity from the radar, the distance-Doppler clustering operation cannot distinguish between the two objects after their point clouds are mapped to the range-Doppler plane. As shown in FIG. 5, a radar captures point clouds of two moving objects, and moving objects A and B have the same distance from the radar and the same velocity relative to the radar. As shown in FIG. 6, after mapping to the range-Doppler plane, the radar point clouds are mixed and result in only one range-Doppler point cloud cluster, making it impossible to distinguish between objects A and B. In contrast, in an embodiment of the present invention, secondary clustering is performed on the range-Doppler point cloud cluster determined to be a moving object by the above-mentioned 102, and point clouds belonging to different moving objects within the range-Doppler point cloud cluster can be distinguished using spatial position information (spatial dimension) or angle information (angular dimension).

Figure 7 shows an example of performing secondary clustering on the range-Doppler point cloud clusters in the spatial dimension (abbreviated as spatial position clustering). As shown in Figure 7, the method includes the following steps:

701: Perform spatial location correction for all points in each range-Doppler point cloud cluster whose range-frequency points are greater than a fourth threshold T _r ;
702: Based on the corrected spatial positions, perform spatial position clustering on all points in each distance-Doppler point cloud cluster, and obtain a spatial point cloud cluster and/or an unclustered point cloud corresponding to each distance-Doppler point cloud cluster as the secondary clustered point cloud cluster.

In an embodiment of the present invention, based on the spatial distance of points, spatial location clustering can divide points belonging to the same distance-Doppler point cloud cluster into several sub-point cloud clusters and a single non-cluster point cloud. In an embodiment of the present invention, the sub-point cloud clusters obtained by spatial location clustering are referred to as spatial point cloud clusters.

In an embodiment of the present invention, spatial position clustering involves two steps: first, correction (701) is performed on the spatial positions (i.e., X, Y, Z) of points that require spatial position correction to resolve the problem of radar measurement uncertainty at long distances; then, clustering is performed using the corrected spatial positions of the points. The present invention is not limited to a clustering method, and conventional clustering methods such as DBSCAN and OPTICS may be adopted.

Figure 8 shows an example of spatial position correction. As shown in Figure 8, the method includes the following steps:

801: input point o _i = (r _i ,v _i ,x _i ,y _i ,z _i ,α _i ,β _i ), where r _i is the distance frequency point of point o _i , v _i is the Doppler velocity frequency point of point o _i , (x _i ,y _i ,z _i ) is the spatial coordinate of point o _i , α _i and β _i are the horizontal and vertical angle frequency points of point o _i respectively;
802: Determine whether the distance frequency point r _i of the point o _i is greater than the threshold T _r , if yes, perform the subsequent operation, if not, do not perform spatial position correction;
803: Perform spatial position correction for points that require spatial position correction;
804: No correction of the spatial position of the point is required;
805: (x', y', z') are the spatial coordinates of the point after correction.

In an embodiment of the present invention, two spatial position correction methods are shown, as shown in 803 of Figure 8.

For example, for each distance frequency point greater than the fourth threshold T _r , keep the value of its spatial coordinate x unchanged, set the value of its spatial coordinate z to a predetermined value, and recalculate the value of its spatial coordinate y;
It is expressed as:

Wherein, (x _i , y _i , z _i ) is the original spatial coordinate of the point, (x _i ', y _i ', z _i ') is the updated spatial coordinate of the point, and z _p is the predetermined value, which is the average height of the moving object relative to the radar.

Furthermore, for example, for each distance frequency point greater than the fourth threshold T _r , the y value and z value of the spatial coordinate are set to predetermined values, and the x value of the spatial coordinate is calculated again.
It is expressed as:

Wherein, (x _i , y _i , z _i ) is the original spatial coordinate of the point, (x _i ', y _i ', z _i ') is the updated spatial coordinate of the point, y _p and z _p are predetermined values, y _p is the average value of the Y axis of the center line relative to the radar in the monitoring scene, and z _p is the average height of the moving object relative to the radar.

Note that the two spatial position correction methods described above are merely examples, and the present invention is not limited to these.

In equations (4) and (5), the value of the spatial coordinate Z is set to a predetermined value _zp , which is the average height of the moving object relative to the radar. In equation (4), the value of the spatial coordinate X is kept unchanged, and then the value of Y is recalculated. In equation (5), the value of the spatial coordinate Y is set to a predetermined value _yp , and then the value of X is recalculated. This correction method is suitable for a long and narrow object tracking scene (e.g., a corridor), where the Y axis is oriented along the length of the long and narrow scene, and the predetermined value _yp can be set to the average value of the Y axis relative to the radar of the center line of the surveillance scene, and the predetermined value _zp is the average height of the moving object relative to the radar.

In an embodiment of the present invention, after spatial position correction is performed on the points in the range-Doppler point cloud cluster, clustering is performed based on the spatial distance of the corrected points to obtain multiple spatial point cloud clusters and unclustered point clouds. A spatial point cloud cluster consists of multiple points that are close to each other in space, where the number of points is greater than a threshold _Tn2 (referred to as the fifth threshold), and the distance from each point to the point cloud cluster is less than a threshold _Td2 (referred to as the sixth threshold).

In some embodiments, the distance from each point in each spatial point cloud cluster to that spatial point cloud cluster is the minimum of the distances from that point to all other points in that spatial point cloud cluster.

For example, D={o _l , 1≦l≦n _D } represents a spatial point cloud cluster, where n _D is the number of points in the spatial point cloud cluster, and n _D >T _n2 ; o _l is the l-th point belonging to the spatial point cloud cluster, and its distance to the spatial point cloud cluster is the minimum value of the distance between the point o _l and other points in the spatial point cloud cluster;
where d(o _j ,o _l ) is the distance in space between points o _j and o _l . As mentioned above, the distance from the point to the spatial point cloud cluster is less than the threshold T _d2 , i.e., d(o _l ,C)<T _d2 .

In some embodiments, the distance between two points in a spatial point cloud cluster is the Euclidean distance between the spatial coordinates of the two points.

For example, the distance between points o _j and o _l is the Euclidean distance between the spatial coordinates of these two points, i.e.,
where (x' _j , y' _j , z' _j ) and (x' _l , y' _l , z' _l ) are the corrected spatial coordinates of points o _j and o _l, respectively.

Note that the above-described method for calculating the distance between two points in a spatial point cloud cluster is merely an example, and the present invention is not limited thereto. In the present invention, the distance between two points in a spatial point cloud cluster may also be calculated using other feasible methods.

Figure 9 shows an example of performing secondary clustering on the range-Doppler point cloud clusters in the angle dimension (called azimuth clustering). As shown in Figure 9, the method includes the following steps:

901: For all points in each range-Doppler point cloud cluster, perform azimuth clustering based on the azimuth angle information of all the points, and obtain an azimuth point cloud cluster and/or an unclustered point cloud corresponding to each range-Doppler point cloud cluster as the secondary clustered point cloud cluster.

In an embodiment of the present invention, based on the angular frequency point distance of the points, azimuth clustering can divide points belonging to the same distance-Doppler point cloud cluster into several sub-point cloud clusters and one non-cluster point cloud. In an embodiment of the present invention, the sub-point cloud cluster obtained by azimuth clustering is called an azimuth point cloud cluster.

In an embodiment of the present invention, one orientation point cloud cluster consists of multiple points with close angular frequency points, among which the number of points is greater than a threshold T _n3 (referred to as the seventh threshold), and the distance from each point to the orientation point cloud cluster is less than a threshold T _d3 (referred to as the eighth threshold).

In some embodiments, the distance from each point in each oriented point cloud cluster to that oriented point cloud cluster is the minimum of the distances from that point to all other points in that oriented point cloud cluster.

For example, E={o _m , 1≦m≦n _E } represents an oriented point cloud cluster, where n _E is the number of points in the oriented point cloud cluster, and n _E <T _n3 ; o _m is the m-th point belonging to the oriented point cloud cluster, and its distance to the oriented point cloud cluster is the minimum distance between point o _m and other points in the oriented point cloud cluster;
where d(o _j ,o _m ) is the distance between points o _j and o _m at the angular frequency point. As mentioned above, the distance from the point to the orientation point cloud cluster is less than the threshold T _d3 , i.e., d(o _m ,E)<T _d3 .

In some embodiments, the distance between two points in an orientation point cloud cluster is a weighted sum of the absolute value of the difference in horizontal angular frequency points between the two points and the absolute value of the difference in vertical angular frequency points between the two points.

For example, the angular frequency point distance between points o _j and o _m is the weighted sum of the absolute value of the horizontal angular frequency point difference and the absolute value of the vertical angular frequency point difference between these two points, i.e.,
d(o _j ,o _m )=b|α _j -α _m |+(1-b)|β _j -β _m | (7)
is.

Wherein, α _j and β _j are the horizontal and vertical angular frequency points of point o _j , α _m and β _m are the horizontal and vertical angular frequency points of point o _m , and b is a predetermined weight value between 0 and 1.

Note that the present invention is not limited to a method for clustering range-Doppler point cloud clusters in the angle dimension, and conventional clustering methods such as DBSCAN and OPTICS may also be used.

In an embodiment of the present invention, 103 can determine the tracked object based on two clustering results (distance-Doppler point cloud clusters obtained by performing range-Doppler clustering processing on radar point clouds, and secondary clustered point cloud clusters obtained by performing secondary clustering on each range-Doppler point cloud cluster). Among them, depending on the secondary clustering method, the secondary clustered point cloud clusters are either spatial point cloud clusters or azimuth point cloud clusters.

In some embodiments, if the number of secondary clustered point cloud clusters corresponding to the distance-Doppler point cloud cluster is greater than 1, the secondary clustered point cloud cluster is the tracking object; and if the number of secondary clustered point cloud clusters corresponding to the distance-Doppler point cloud cluster is not greater than 1, the distance-Doppler point cloud cluster is the tracking object.

Figure 10 shows an example of determining a tracking object. As shown in Figure 10, the method includes the following steps:

1001: Input a range-Doppler point cloud cluster and its corresponding secondary clustered point cloud cluster, where C represents one range-Doppler point cloud cluster and {S _i , 0≦i≦n _s } represents its corresponding secondary clustered point cloud cluster;
1002: Determine whether the number of secondary clustering point cloud clusters n _s is greater than 1, if yes, execute 1004, if not, execute 1003;
1003: Determine the distance-Doppler point cloud cluster C as the tracking object;
1004: Determine the secondary clustered point cloud cluster as the tracked object.

Note that while only the operations or processes related to the present invention have been described above, the present invention is not limited to these. The method may also include other operations or processes, and reference can be made to related art for the specific content of these operations or processes.

The method in this embodiment of the present invention improves tracking accuracy and extends the effective tracking range through a two-step clustering strategy, solving the problem of poor radar spatial measurement accuracy at long distances, which makes point clouds prone to dispersion and makes clustering impossible.

<Example of the second aspect>
In the embodiment of the present invention, a noise removal method is provided, which is similar to the noise removal method of the embodiment of the first aspect of the present invention shown in Fig. 3, and therefore a redundant description of the same content will be omitted.

Figure 11 shows an example of a noise removal method in an embodiment of the present invention. As shown in Figure 11, the method includes the following steps:

1101: Performing statistics on point cloud noise information on the range-Doppler plane, the point cloud noise information including a noise statistical total number n _n and a number n _r,v of noise points appearing at each position on the range-Doppler plane;
1102: Analyzing the point cloud clusters on the range-Doppler plane based on the point cloud noise information to determine whether the point cloud clusters are noise-induced point cloud clusters;
1103: If the point cloud cluster is a noisy point cloud cluster, remove the noisy point cloud cluster in the range-Doppler plane.

In 1102, in some implementations, obtain a probability P(C) that each range-Doppler point cloud cluster C is noise based on the point cloud noise information, and if the probability P(C) that the point cloud cluster C is noise is greater than a third threshold _Tn , the point cloud cluster C is deemed to be a point cloud cluster caused by noise.

In the above embodiment, the probability P(C) that each point cloud cluster C is noise is the average probability that each point in the point cloud cluster C is noise, and the probability that each point in the point cloud cluster C is noise is the ratio of the number of times n _r,v that a noise point appears at each position on the range-Doppler plane to the total number of times n _n in the noise statistics.

In an embodiment of the present invention, as shown in FIG. 11, the method further includes the following steps:

1104: Update the point cloud noise information on the range-Doppler plane, which includes adding 1 to the noise statistics total count and adding 1 to the count of noise occurrences at positions on the range-Doppler plane corresponding to all points belonging to the range-Doppler noise point cloud cluster and all points in the non-cluster point cloud.

For specific update methods, please refer to the explanation of Figure 4 in the first aspect of the embodiment.

Note that while the above describes only the operations or processes related to the present invention, the present invention is not limited to these. The method may also include other operations or processes, and reference can be made to related art for the specific content of these operations or processes.

According to the method of the present invention, based on point cloud noise information on the range-Doppler plane, it is possible to determine whether a point cloud cluster on the range-Doppler plane is a point cloud cluster caused by noise, and then remove the point cloud cluster caused by noise, thereby further improving tracking accuracy.

<Example of the third aspect>
In the embodiment of the present invention, a radar tracking device is provided, and the radar tracking device corresponds to the radar tracking method of the first aspect of the embodiment, so that the overlapping description of the same content will be omitted.

Figure 12 is a diagram illustrating an example of a radar tracking device in an embodiment of the present invention. As shown in Figure 12, the radar tracking device 1200 in an embodiment of the present invention includes the following:

A first clustering unit 1201: performs clustering on the point cloud acquired by radar detection in the range-Doppler plane to obtain a range-Doppler point cloud cluster;
A second clustering unit 1202: performs secondary clustering on the range-Doppler point cloud cluster in a spatial dimension or an angular dimension to obtain a secondary clustered point cloud cluster;
A determining unit 1203: determines a tracking object based on the range-Doppler point cloud cluster and its corresponding secondary clustered point cloud cluster.

In some embodiments, the first clustering unit 1201 maps the point cloud onto a range-Doppler plane, and divides the point cloud into a plurality of point cloud clusters and non-clustered point clouds based on the distance of the point cloud on the range-Doppler plane, where the number of points in each point cloud cluster is greater than a first threshold T _n1 and the distance from each point in each point cloud cluster to the point cloud cluster is less than a second threshold T _d1 . Specific clustering methods include, but are not limited to, DBSCAN, OPTICS, etc.

In the above example, the distance from each point in each point cloud cluster to that point cloud cluster is the minimum of the distances from that point to all other points in that point cloud cluster.

In the above example, the distance between two points in a point cloud cluster is a weighted sum of the absolute value of the difference in the distance frequency points of the two points and the absolute value of the difference in the Doppler velocity frequency points of the two points.

For example, the distance between points within a point cloud cluster can be calculated using the following formula:

d(o _j ,o _k )=a|r _j -r _k |+(1-a)|v _j -v _k |
Wherein, r _j and r _k are the distance frequency points of points o _j and o _k in the point cloud cluster C, respectively, v _j and v _k are the Doppler velocity frequency points of points o _j and o _k in the point cloud cluster C, respectively, and a is a predetermined weight value between 0 and 1.

In an embodiment of the present invention, as shown in FIG. 12, the device 1200 optionally further includes:

Noise removal unit 1204: Removes point cloud clusters due to noise within the range-Doppler point cloud clusters.

In some embodiments, the noise removal unit 1204 performs statistics on point cloud noise information on the range-Doppler plane, where the point cloud noise information includes a noise statistical total number n _n and a number n _r,v of noise points appearing at each position on the range-Doppler plane; analyzes the range-Doppler point cloud cluster based on the point cloud noise information, determines whether the range-Doppler point cloud cluster is a noise-caused point cloud cluster; and if the range-Doppler point cloud cluster is a noise-caused point cloud cluster, removes the noise-caused point cloud cluster on the range-Doppler plane.

In some embodiments, the noise removal unit 1204 obtains a probability P(C) that each range-Doppler point cloud cluster C is noise based on the point cloud noise information, and if the probability P(C) that the range-Doppler point cloud cluster C is noise is greater than a third threshold _Tn , the range-Doppler point cloud cluster C is deemed to be a point cloud cluster caused by noise.

In some embodiments, the probability P(C) of each of the range-Doppler point cloud clusters C being noise is the average probability of each point in the range-Doppler point cloud cluster C being noise, and the probability of each point in the range-Doppler point cloud cluster C being noise is the ratio of the number of times n _r,v that a noise point appears at each position on the range-Doppler plane to the total number of times n _n in the noise statistics.

In some embodiments, the noise removal unit 1204 may further update the point cloud noise information in the range-Doppler plane by adding 1 to the noise statistics total count and adding 1 to the count of noise occurrences at range-Doppler plane positions corresponding to all points belonging to the range-Doppler noise point cloud cluster and all points in the non-cluster point cloud.

In some embodiments, the second clustering unit 1202 performs secondary clustering on the distance-Doppler point cloud clusters in the spatial dimension by performing spatial position correction on all points in each distance-Doppler point cloud cluster whose distance frequency points are greater than a fourth threshold T _r , and performing spatial position clustering on all points in each distance-Doppler point cloud cluster according to the corrected spatial positions, to obtain spatial point cloud clusters and/or unclustered point clouds as the secondary clustered point cloud clusters corresponding to each distance-Doppler point cloud cluster. Specific clustering methods include, but are not limited to, DBSCAN, OPTICS, etc.

In the above embodiment, the spatial position correction includes, for each point whose distance frequency point is greater than the fourth threshold T _r , keeping the value of its spatial coordinate x unchanged, setting the value of its spatial coordinate z to a predetermined value, and recalculating the value of its spatial coordinate y;
It may also be expressed as:

Wherein, (x _i , y _i , z _i ) is the original spatial coordinate of the point, (x _i ', y _i ', z _i ') is the updated spatial coordinate of the point, and z _p is the predetermined value, which is the average height of the moving object relative to the radar.

In some embodiments, the spatial position correction includes, for each distance frequency point greater than a fourth threshold T _r , setting the y value and z value of the spatial coordinate to predetermined values, and recalculating the x value of the spatial coordinate;
It may also be expressed as:

Wherein, (x _i , y _i , z _i ) is the original spatial coordinate of the point, (x _i ', y _i ', z _i ') is the updated spatial coordinate of the point, y _p and z _p are predetermined values, y _p is the average value of the Y axis of the center line relative to the radar in the monitored scene, and z _p is the average height of the moving object relative to the radar.

In some embodiments, the number of points in each spatial point cloud cluster is greater than a fifth threshold T _n2 and the distance from each point in each spatial point cloud cluster to the spatial point cloud cluster is less than a sixth threshold T _d2 .

In some embodiments, the distance from each point in each spatial point cloud cluster to that spatial point cloud cluster is the minimum of the distances from that point to all other points in that spatial point cloud cluster.

In the above example, the distance between two points in a spatial point cloud cluster is the Euclidean distance between the spatial coordinates of the two points.

In some embodiments, the second clustering unit 1202 performs secondary clustering on the range-Doppler point cloud clusters in the angle dimension by performing azimuth clustering on all points in each range-Doppler point cloud cluster based on the azimuth angle information of all the points, and obtaining an azimuth point cloud cluster and/or an unclustered point cloud as the secondary clustered point cloud cluster for each range-Doppler point cloud cluster. Specific clustering methods include, but are not limited to, DBSCAN, OPTICS, etc.

In some embodiments, the number of points in each oriented point cloud cluster is greater than a seventh threshold T _n3 and the distance from each point in each oriented point cloud cluster to the oriented point cloud cluster is less than an eighth threshold T _d3 .

In some embodiments, the distance from each point in each oriented point cloud cluster to that point cloud cluster is the minimum of the distances from that point to all other points in that oriented point cloud cluster.

In the above example, the distance between two points in an orientation point cloud cluster is the weighted sum of the absolute value of the difference in horizontal angular frequency points between the two points and the absolute value of the difference in vertical angular frequency points between the two points.

For example, the distance between points in an oriented point cloud cluster can be calculated using the following formula:

d(o _j ,o _m )=b|α _j -α _m |+(1-b)|β _j -β _m |
Wherein, α _j and β _j are the horizontal and vertical angular frequency points of point o _j in the orientation point cloud cluster, α _m and β _m are the horizontal and vertical angular frequency points of point o _m in the orientation point cloud cluster, and b is a predetermined weight value between 0 and 1.

In some embodiments, if the number of secondary clustered point cloud clusters corresponding to the range-Doppler point cloud cluster is greater than 1, the determination unit 1203 determines the secondary clustered point cloud cluster as the tracking object; if the number of secondary clustered point cloud clusters corresponding to the range-Doppler point cloud cluster is not greater than 1, the determination unit 1203 determines the range-Doppler point cloud cluster as the tracking object.

Note that while only the components or modules related to the present invention have been described, the present invention is not limited to these. The radar tracking device 1200 may also include other components or modules, and reference can be made to related art for the specific content of these components or modules.

According to an embodiment of the present invention, a two-step clustering strategy improves tracking accuracy and extends the effective tracking range, thereby solving the problem of poor radar spatial measurement accuracy at long distances, which makes point clouds prone to dispersion and makes clustering impossible.

<Example of the fourth aspect>
An embodiment of the present invention provides a noise removal device, which corresponds to the noise removal method of the second aspect of the embodiment, and redundant description of the same content will be omitted.

Figure 13 shows an example of a noise removal device according to an embodiment of the present invention. As shown in Figure 13, a noise removal device 1300 according to an embodiment of the present invention includes the following:

A statistical unit 1301 performs statistical analysis on point cloud noise information in the range-Doppler plane, the point cloud noise information including the total number of noise statistics n _n and the number of times that a noise point appears at each position in the range-Doppler plane n _r,v ;
An analysis unit 1302: analyzes the point cloud clusters in the range-Doppler plane according to the point cloud noise information, and determines whether the range-Doppler point cloud clusters are point cloud clusters caused by noise;
Processing unit 1303: if the range-Doppler point cloud cluster is a noisy point cloud cluster, processing unit 1303 removes the noisy point cloud cluster in the range-Doppler plane.

In some embodiments, the analysis unit 1302 obtains a probability P(C) that each range-Doppler point cloud cluster C is noise based on the point cloud noise information, and if the probability P(C) that the range-Doppler point cloud cluster C is noise is greater than a third threshold _Tn , the range-Doppler point cloud cluster C is deemed to be a point cloud cluster caused by noise.

In some embodiments, the probability P(C) of each of the range-Doppler point cloud clusters C being noise is the average probability of each point in the range-Doppler point cloud cluster C being noise, and the probability of each point in the range-Doppler point cloud cluster C being noise is the ratio of the number of times n _r,v that a noise point appears at each position on the range-Doppler plane to the total number of times n _n in the noise statistics.

In some embodiments, as shown in FIG. 13, the device 1300 further includes:

Update unit 1304: Updates the point cloud noise information in the range-Doppler plane, which adds 1 to the total number of noise statistics and adds 1 to the number of times noise appears at positions in the range-Doppler plane corresponding to all points belonging to the range-Doppler noise point cloud cluster and all points in the non-cluster point cloud.

Note that while only the components or modules related to the present invention have been described above, the present invention is not limited to these. The noise removal device 1300 may also include other components or modules, and reference can be made to related art for the specific content of these components or modules.

According to an embodiment of the present invention, it is possible to further improve tracking accuracy by determining whether a point cloud cluster on the range-Doppler plane is a point cloud cluster caused by noise based on point cloud noise information on the range-Doppler plane, and then removing the point cloud cluster caused by noise.

<Example of the fifth aspect>
In an embodiment of the present invention, a computing device is provided, and the computing device may be, for example, a computer, a server, a workstation, a desktop computer, a smartphone, etc., but the embodiment of the present invention is not limited thereto.

Figure 14 is a diagram showing a computer device in an embodiment of the present invention. As shown in Figure 14, computer device 1400 includes at least one interface (not shown), a processor (e.g., a central processing unit (CPU)) 1401, and a memory 1402, with memory 1402 connected to processor 1401. Memory 1402 can store various data and also store program 1403, and execute program 1403 under the control of processor 1401 to store various data, such as predetermined thresholds, predetermined conditions, etc.

In some embodiments, the functionality of the radar tracking device 1200 described in the embodiments of the third aspect may be integrated into a processor 1401 to implement the radar tracking method described in the embodiments of the first aspect. For example, the processor 1401 may be configured to perform the following processes:
Perform clustering on the point cloud acquired by radar detection in the range-Doppler plane to obtain a range-Doppler point cloud cluster;
performing secondary clustering on the range-Doppler point cloud clusters in a spatial dimension or an angular dimension to obtain secondary clustered point cloud clusters;
A tracked object is determined based on the range-Doppler point cloud cluster and its corresponding secondary clustered point cloud cluster.

In some embodiments, the functionality of the denoising device 1300 of the embodiments of the fourth aspect can be integrated into a processor 1401 to implement the denoising method described in the embodiments of the second aspect. For example, the processor 1401 may be configured to perform the following processes:
Perform statistics on point cloud noise information on the range-Doppler plane, where the point cloud noise information includes a noise statistical total number n _n and a number n _r,v of noise points appearing at each position on the range-Doppler plane;
Analyzing the point cloud clusters on the range-Doppler plane based on the point cloud noise information to determine whether the range-Doppler point cloud clusters are point cloud clusters caused by noise;
If the range-Doppler point cloud cluster is a point cloud cluster due to noise, the point cloud cluster due to noise in the range-Doppler plane is removed.

In some embodiments, the radar tracking device 1200 described in the embodiments of the third aspect or the noise reduction device 1300 described in the embodiments of the fourth aspect may be arranged separately from the processor 1401. For example, the radar tracking device 1200 or the noise reduction device 1300 may be configured as a chip connected to the processor 1401, and the functions of the radar tracking device 1200 or the noise reduction device 1300 may be realized under the control of the processor 1401.

Note that the computer device 1400 may further include a display 1405 and an I/O device 1404, or may not necessarily include all of the components shown in FIG. 14, and may further include, for example, a camera head and/or radar (not shown) for acquiring images or radar point clouds, or the computer device 1400 may further include components not shown in FIG. 14, for which reference may be made to the prior art.

In embodiments of the present invention, the processor 1401 may be referred to as a controller or operational control and may include a microprocessor or other processing and/or logic device, and the processor 1401 may receive inputs and control the operation of the components of the computing device 1400.

In an embodiment of the present invention, the storage device 1402 may include, for example, one or more of a buffer, flash memory, HDD, removable medium, volatile storage, non-volatile storage, or other suitable device, and may store various types of information and may further store programs for information processing. The processor 1401 may also execute the programs stored in the storage device 1402 to store, process, and otherwise process information. Since the functions of the other components are similar to those of conventional components, detailed descriptions thereof will be omitted here. Each component of the computer device 1400 may be implemented using dedicated hardware, firmware, software, or a combination thereof, all of which fall within the scope of the present invention.

An embodiment of the present invention further provides a computer-readable program, which, when executed in a computing device, causes the computing device to perform a method according to the first or second aspect of the embodiment.

An embodiment of the present invention further provides a storage medium storing a computer-readable program, wherein the computer-readable program executes the method described in the first or second aspect of the embodiment in a computer device.

The above-described devices and methods of the present invention may be realized by hardware or a combination of hardware and software. The present invention also relates to a computer-readable program that, when executed by a logic component, causes the logic component to implement the above-described devices or components, or to perform the various methods or steps described above. The present invention also relates to a storage medium capable of storing such a program, such as a hard disk, magnetic disk, optical disk, DVD, flash memory, etc.

The object of the present invention may also be realized in the following manner: a storage medium on which the above-mentioned executable program code is stored is provided directly or indirectly to a system or device, and a computer or central processing unit (CPU) in the system or device reads and executes the above-mentioned program code. In this case, as long as the system or device has the ability to execute a program, the embodiment of the present invention is not limited to a program; the program may be in any format, such as an object-oriented program, a program executed by an interpreter, or a script program provided to an operating system.

These machine-readable storage media may include, but are not limited to, various types of memory and storage units, semiconductor devices, optical, magnetic, and magnetic disks such as magneto-optical disks, and other media suitable for storing information.

Alternatively, a computer can connect to a corresponding website on the Internet, download and install the computer program code according to the present invention onto the computer, and then run the program to realize the technical solution of the present invention.

The present invention also provides a program product including machine-readable instruction code. When such instruction code is read and executed by a machine, it is capable of executing the method of the above-described embodiment of the present invention. Accordingly, various storage media that carry such program products, such as magnetic disks (including floppy disks (registered trademark)), optical disks (including CD-ROMs and DVDs), magneto-optical disks (including MDs (registered trademark)), and semiconductor memory devices, are also included in the present invention.

The above-mentioned storage media may include, for example, magnetic disks, optical disks, magneto-optical disks, semiconductor memory devices, etc., but are not limited to these.

Furthermore, each operation (process) in the above-described method can also be implemented in the form of a computer-executable program stored on various machine-readable storage media.

In addition, the following supplementary notes are provided regarding the above examples.

(Appendix 1)
1. A radar tracking method comprising:
S1: Perform clustering on the point cloud acquired by radar detection in the range-Doppler plane to obtain a range-Doppler point cloud cluster;
S3: Perform secondary clustering on the range-Doppler point cloud cluster in a spatial dimension or an angular dimension to obtain a secondary clustered point cloud cluster; and
S4: determining a tracked object based on the range-Doppler point cloud cluster and its corresponding secondary clustered point cloud cluster.

(Appendix 2)
2. The method of claim 1, comprising:
S1 is
S11: Mapping the point cloud onto a range-Doppler plane;
S12: Dividing the point cloud into a plurality of point cloud clusters and non-clustered point clouds based on the distance of the point cloud in the distance-Doppler plane, the number of points in each point cloud cluster is greater than a first threshold T _n1 , and the distance from each point in each point cloud cluster to the point cloud cluster is less than a second threshold T _d1 .

(Appendix 3)
10. The method of claim 2,
A method wherein the distance from each point in each point cloud cluster to the point cloud cluster is the minimum of the distances from that point to all other points in the point cloud cluster.

(Appendix 4)
4. The method of claim 3,
The method, wherein the distance between two points in a point cloud cluster is a weighted sum of the absolute value of the difference in distance frequency points of the two points and the absolute value of the difference in Doppler velocity frequency points of the two points.

(Appendix 5)
5. The method of claim 4,
The distance between points in a point cloud cluster is
d(o _j ,o _k )=a|r _j -r _k |+(1-a)|v _j -v _k |
is calculated based on
wherein r _j and r _k are the distance frequency points of points o _j and o _k in the point cloud cluster C, respectively, v _j and v _k are the Doppler velocity frequency points of points o _j and o _k in the point cloud cluster C, respectively, and a is a predetermined weight value between 0 and 1.

(Appendix 6)
The method of claim 1, further comprising:
S2: Removing noisy point cloud clusters in the range-Doppler point cloud clusters.

(Appendix 7)
7. The method of claim 6,
S2 is
S21: perform statistics on point cloud noise information on the range-Doppler plane, the point cloud noise information including: a total noise statistical count n _n and a number of times n _r,v that a noise point appears at each position on the range-Doppler plane;
S22: Analyzing the range-Doppler point cloud cluster based on the point cloud noise information, and determining whether the range-Doppler point cloud cluster is a point cloud cluster caused by noise; and
S23: If the range-Doppler point cloud cluster is a noisy point cloud cluster, removing the noisy point cloud cluster in the range-Doppler plane.

(Appendix 8)
8. The method of claim 7,
In S22,
determining a probability P(C) that each range-Doppler point cloud cluster C is noise based on the point cloud noise information; and if the probability P(C) that the range-Doppler point cloud cluster C is noise is greater than a third threshold _Tn , determining that the range-Doppler point cloud cluster C is a noise point cloud cluster.

(Appendix 9)
9. The method of claim 8,
The probability P(C) that each of the range-Doppler point cloud clusters C is noise is the average probability that each point in the range-Doppler point cloud cluster C is noise,
The method wherein the probability that each point in the range-Doppler point cloud cluster C is noise is the ratio of the number of times n _r,v that a noise point appears at each position in the range-Doppler plane to the total number of times n _n in the noise statistics.

(Appendix 10)
8. The method of claim 7,
S2 further
S24: Updating point cloud noise information in the range-Doppler plane;
S24 is
adding 1 to the noise statistics total count; and adding 1 to the noise occurrence counts at range-Doppler plane locations corresponding to all points in the range-Doppler noise point cloud cluster and all points in the non-cluster point cloud.

(Appendix 11)
2. The method of claim 1, comprising:
In S3, performing secondary clustering on the range-Doppler point cloud clusters in the spatial dimension is
S31: Perform spatial location correction for all points in each range-Doppler point cloud cluster whose range-frequency points are greater than a fourth threshold T _r ; and
S32: performing spatial position clustering on all points in each range-Doppler point cloud cluster based on the corrected spatial positions, and obtaining a spatial point cloud cluster and/or an unclustered point cloud corresponding to each range-Doppler point cloud cluster as the secondary clustered point cloud cluster.

(Appendix 12)
12. The method of claim 11,
In S31, for each point whose distance frequency point is greater than a fourth threshold T _r , the value of the spatial coordinate x is kept unchanged, the value of the spatial coordinate z is set to a predetermined value, and the value of the spatial coordinate y is recalculated;
and
wherein (x _i ,y _i ,z _i ) is the original spatial coordinate of the point, (x _i ',y _i ',z _i ') is the updated spatial coordinate of the point, and z _p is the predetermined value, which is the average height of the moving object relative to the radar.

(Appendix 13)
12. The method of claim 11,
In step S31, for each distance frequency point greater than the fourth threshold T _r , the y value and z value of the spatial coordinate are set to predetermined values, and the x value of the spatial coordinate is calculated again;
and
wherein (x _i , y _i , z _i ) is the original spatial coordinate of the point, (x _i ', y _i ', z _i ') is the updated spatial coordinate of the point, y _p and z _p are predetermined values, y _p is the average value of the Y axis of the center line relative to the radar in the monitoring scene, and z _p is the average height of the moving object relative to the radar.

(Appendix 14)
12. The method of claim 11,
In S32, the number of points in each spatial point cloud cluster is greater than a fifth threshold T _n2 and the distance from each point in each spatial point cloud cluster to the spatial point cloud cluster is less than a sixth threshold T _d2 .

(Appendix 15)
15. The method of claim 14,
A method wherein the distance from each point in each spatial point cloud cluster to the spatial point cloud cluster is the minimum of the distances from the point to all other points in the spatial point cloud cluster.

(Appendix 16)
16. The method of claim 15,
The method, wherein the distance between two points in a spatial point cloud cluster is the Euclidean distance of the spatial coordinates of the two points.

(Appendix 17)
2. The method of claim 1, comprising:
In S3, performing secondary clustering on the range-Doppler point cloud clusters in the angle dimension is
S31': A method including: performing azimuth clustering for all points in each range-Doppler point cloud cluster based on azimuth angle information of all the points, and obtaining an azimuth point cloud cluster and/or an unclustered point cloud corresponding to each range-Doppler point cloud cluster as the secondary clustered point cloud cluster.

(Appendix 18)
18. The method of claim 17,
The number of points in each oriented point cloud cluster is greater than a seventh threshold T _n3 and the distance from each point in each oriented point cloud cluster to the oriented point cloud cluster is less than an eighth threshold T _d3 .

(Appendix 19)
19. The method of claim 18,
A method wherein the distance from each point in each oriented point cloud cluster to that point cloud cluster is the minimum of the distances from that point to all other points in that oriented point cloud cluster.

(Appendix 20)
19. The method of claim 18,
The method, wherein the distance between two points in an orientation point cloud cluster is a weighted sum of the absolute value of the difference in horizontal angular frequency points of the two points and the absolute value of the difference in vertical angular frequency points of the two points.

(Appendix 21)
21. The method of claim 20,
The distance between points in an oriented point cloud cluster is
d(o _j ,o _m )=b|α _j -α _m |+(1-b)|β _j -β _m |
is calculated based on
wherein α _j and β _j are the horizontal and vertical angular frequency points of point o _j in the orientation point cloud cluster, α _m and β _m are the horizontal and vertical angular frequency points of point o _m in the orientation point cloud cluster, and b is a predetermined weight value between 0 and 1.

(Appendix 22)
2. The method of claim 1, comprising:
S4 is
if the number of secondary clustered point cloud clusters corresponding to the distance-Doppler point cloud cluster is greater than 1, the secondary clustered point cloud cluster is set as a tracking object; and if the number of secondary clustered point cloud clusters corresponding to the distance-Doppler point cloud cluster is not greater than 1, the distance-Doppler point cloud cluster is set as a tracking object.

(Appendix 23)
A noise removal method, comprising:
S21: perform statistics on point cloud noise information on the range-Doppler plane, the point cloud noise information including: a total noise statistical count n _n and a number of times n _r,v that a noise point appears at each position on the range-Doppler plane;
S22: Analyzing the point cloud clusters in the range-Doppler plane based on the point cloud noise information to determine whether the range-Doppler point cloud clusters are point cloud clusters caused by noise; and
S23: If the range-Doppler point cloud cluster is a noisy point cloud cluster, removing the noisy point cloud cluster in the range-Doppler plane.

(Appendix 24)
24. The method of claim 23,
In S22,
According to the point cloud noise information, obtain a probability P(C) that each range-Doppler point cloud cluster C is noise;
If the probability P(C) that the range-Doppler point cloud cluster C is noise is greater than a third threshold _Tn , the range-Doppler point cloud cluster C is considered to be a point cloud cluster due to noise.

(Appendix 25)
25. The method of claim 24,
The probability P(C) that each of the range-Doppler point cloud clusters C is noise is the average probability that each point in the range-Doppler point cloud cluster C is noise;
The method wherein the probability that each point in the range-Doppler point cloud cluster C is noise is the ratio of the number of times n _r,v that a noise point appears at each position in the range-Doppler plane to the total number of times n _n in the noise statistics.

(Appendix 26)
24. The method of claim 23, further comprising:
S24: updating point cloud noise information in the range-Doppler plane;
S24 is
adding 1 to the noise statistics total count; and adding 1 to the noise application counts at distance-Doppler plane positions corresponding to all points in distance-Doppler noise point cloud clusters and all points in non-cluster point clouds.

(Appendix 27)
A computing device including a memory and a processor,
27. A computing device, the storage device storing a computer program, the processor configured to execute the computer program to perform the method of any one of claims 1 to 26.

(Appendix 28)
A storage medium storing a computer-readable program,
A storage medium, the computer readable program causing a computer to execute the method according to any one of appendices 1 to 26 in a computing device.

The above describes a preferred embodiment of the present invention, but the present invention is not limited to this embodiment, and any modifications to the present invention that do not depart from the spirit of the present invention fall within the technical scope of the present invention.

Claims (5)

1. A radar tracking device, comprising:
a first clustering unit that performs clustering on the point cloud obtained by radar detection in the range-Doppler plane to obtain range-Doppler point cloud clusters;
a second clustering unit for performing secondary clustering on the range-Doppler point cloud clusters in a spatial dimension to obtain secondary clustered point cloud clusters; and a determining unit for determining a tracking object based on the range-Doppler point cloud clusters and their corresponding secondary clustered point cloud clusters ,
The second clustering unit performs secondary clustering on the range-Doppler point cloud clusters in the spatial dimension,
Performing spatial location correction for all points in each range-Doppler point cloud cluster whose range-frequency points are greater than a fourth threshold; and
performing spatial location clustering on all points in each range-Doppler point cloud cluster based on the corrected spatial locations, and obtaining a spatial point cloud cluster and/or an unclustered point cloud corresponding to each range-Doppler point cloud cluster as the secondary clustered point cloud cluster;
The unclustered point cloud refers to a point cloud that is not a cluster . 2. The radar tracking device of claim 1,
the first clustering unit maps the point cloud onto a range-Doppler plane, and divides the point cloud into a plurality of point cloud clusters and unclustered point clouds based on the distances of the point clouds on the range- Doppler plane, wherein the number of points in each point cloud cluster is greater than a first threshold, and the distance from each point in each point cloud cluster to the point cloud cluster is less than a second threshold. 2. The radar tracking device of claim 1,
The radar tracking device further includes a denoising unit for removing point cloud clusters due to noise in the range-Doppler point cloud clusters. 2. The radar tracking device of claim 1,
If the number of secondary clustered point cloud clusters corresponding to the distance-Doppler point cloud cluster is greater than 1, the determining unit determines the secondary clustered point cloud cluster as a tracking object;
If the number of secondary clustered point cloud clusters corresponding to the range-Doppler point cloud cluster is not greater than 1, the determination unit determines the range-Doppler point cloud cluster as a tracking object. On the computer,
Perform clustering on the point cloud obtained by radar detection in the range-Doppler plane to obtain range-Doppler point cloud clusters;
1. A program for executing a radar tracking method, the program comprising: performing secondary clustering on the range-Doppler point cloud clusters in a spatial dimension to obtain secondary clustered point cloud clusters; and determining a tracking object based on the range-Doppler point cloud clusters and their corresponding secondary clustered point cloud clusters ,
performing a second-order clustering on the range-Doppler point cloud clusters in the spatial dimension,
Performing spatial location correction for all points in each range-Doppler point cloud cluster whose range-frequency points are greater than a fourth threshold; and
performing spatial location clustering on all points in each range-Doppler point cloud cluster based on the corrected spatial locations, and obtaining a spatial point cloud cluster and/or an unclustered point cloud corresponding to each range-Doppler point cloud cluster as the secondary clustered point cloud cluster;
The unclustered point cloud refers to a point cloud that is not a cluster .