阅读说明：本技术 一种基于毫米波雷达的3d多目标跟踪处理方法 (3D multi-target tracking processing method based on millimeter wave radar ) 是由 苏涛 刘馨璐 杨天园 于 2021-07-21 设计创作，主要内容包括：本发明属于雷达信号处理技术领域,公开了一种基于毫米波雷达的3D多目标跟踪处理方法,先通过毫米波雷达基带信号处理得到空间上多目标的点迹,选取信噪比较强的点迹绘制成3D点云图,根据点云图进行后续的聚类和跟踪的处理；聚类处理的输入是多个3D点云,具有空间三维方向的信息,遍历搜索距离位置邻近且同一类别的点迹,通过同一类的标记得到同类的点迹即为聚类得到的结果,确定该一类点迹的空间位置等参数后输入轨迹跟踪中,进行轨迹算法的点迹关联和点迹管理,得到每一帧的点迹的轨迹,跟踪目标并且标记该目标的ID号。可以实现在复杂的环境中,精确的检测并跟踪多目标,并从三维的空间中,确定目标在空间的位置。(The invention belongs to the technical field of radar signal processing, and discloses a millimeter wave radar-based 3D multi-target tracking processing method, which comprises the steps of firstly obtaining multi-target point traces on a space through millimeter wave radar baseband signal processing, selecting the point traces with strong signal-to-noise ratio to draw a 3D point cloud picture, and carrying out subsequent clustering and tracking processing according to the point cloud picture; the input of the clustering processing is a plurality of 3D point clouds with information in a spatial three-dimensional direction, the point traces which are adjacent to the searching distance position and are in the same category are traversed, the point traces in the same category are obtained through the marks in the same category, namely the result obtained through clustering, the parameters such as the spatial position of the point traces in the category are determined and then input into the track tracking, the point trace correlation and the point trace management of the track algorithm are carried out, the track of the point trace of each frame is obtained, the target is tracked, and the ID number of the target is marked. The method can realize accurate detection and tracking of multiple targets in a complex environment, and determine the position of the target in the space from the three-dimensional space.)

1. A3D multi-target tracking processing method based on millimeter wave radar is characterized by comprising the following steps:

step 1, a millimeter wave radar transmits electromagnetic waves and receives baseband echo signals, and the baseband echo signals are processed to obtain data information of a plurality of traces in space; the data information of each trace point comprises three-dimensional space position and speed information of the trace point;

step 2, clustering the plurality of traces through a clustering algorithm, and drawing the plurality of traces belonging to the same class into a 3D point cloud so as to obtain a plurality of target point clouds; removing noise point clouds from a plurality of target point clouds to obtain effective target point clouds;

step 3, determining an approximate space position and a speed average value V of a target according to the three-dimensional space position and speed information of the trace in the effective target point cloud; wherein the approximate spatial position of the target comprises a position of the target in the direction of the azimuthal X coordinate axisPosition of target in direction of orientation Y coordinate axisThe position of the target in the direction perpendicular to the Z coordinate axis is Z_max；

Step 4, according to the position of the target in the direction of the X coordinate axis of the azimuthPosition of target in direction of orientation Y coordinate axisCalculating to obtain a horizontal distance R and a horizontal angle A, and determining a track tracking route of the target according to the horizontal distance R, the horizontal angle A and the speed average value V;

and 5, carrying out track management on the track tracking route of the target, and updating the track state at any time.

2. The millimeter wave radar-based 3D multi-target tracking processing method according to claim 1, wherein the step 2 specifically comprises the following substeps:

substep 2.1, the clustering algorithm firstly needs to perform traversal marking and category marking on each trace point, the default is 0, secondly needs to distinguish each trace point, and sets the spatial position information of the first point as (x)₀,y₀,z₀) Firstly, marking the point as detected, traversing the whole space by taking the point as a core point, and searching all similar points of the core point to obtain a plurality of points of the same type, which are called as point clouds of the target;

substep 2.2, setting the minimum clustering point number minPointsInCluster in one class, and if the number of the point traces in the same class is greater than the minimum clustering point number minPointsInCluster, determining the point traces as effective target point clouds; otherwise, it is noise point cloud.

3. The millimeter wave radar-based 3D multi-target tracking processing method according to claim 2, wherein in the sub-step 2.1, the method for finding all the homogeneous points of the core points comprises the sub-steps of:

substep 2.1.1, determining core points (x)₀,y₀,z₀) Whether all points in the domain of (1) are homogeneous or not, the formula is:

wherein epsilon is the radius of the clustering algorithm field; v. of₀For the speed of the current core point, yFactor is a weighting factor for setting a Y coordinate, zFactor is a weighting factor for setting a Z coordinate, and vFactor is a weighting factor for setting a speed; (x, y, z) is three-dimensional spatial position information of any one point in the core point region, and v is velocity information of any one point in the core point region;

substep 2.1.2, statistics belong to core point (x)₀,y₀,z₀) Belongs to the field ofThe number of the traces in the same type, numInEps, if the number satisfies numInEps>Traversing and marking the trace points meeting the conditions under the minPointsInCluster condition;

and sequentially taking the traversably marked point traces as core points, and performing traversal search on the points which are not traversably marked until all the marked points are searched to obtain a plurality of points in the same category.

4. The millimeter wave radar-based 3D multi-target tracking processing method according to claim 1, wherein in step 3, the position of the target in the direction of the orientation X coordinate axisComprises the following steps:x_sumthe point clouds are all sums in the X direction of the point clouds in the same category, and N is the total number of the point traces in the category;

the position of the target in the direction of the orientation Y coordinate axisComprises the following steps:y_sumall sums in the Y direction of the point clouds in the same category;

the position of the target in the direction vertical to the Z coordinate axis is Z_maxThe maximum value of the point clouds in the same category in the direction vertical to the Z coordinate axis;

the speed mean value V is: v ═ V_sum/N，v_sumAll velocity sums for the same category of point clouds.

5. The millimeter wave radar-based 3D multi-target tracking processing method according to claim 4, wherein in step 4, the horizontal distance R is:the horizontal angle A is as follows:

6. the millimeter wave radar-based 3D multi-target tracking processing method according to claim 1, wherein in step 4, the determining a trajectory tracking route of the target according to the horizontal distance R, the horizontal angle A and the speed mean V specifically comprises:

representing the moving state of the whole target by using a horizontal distance R, a horizontal angle A and a speed mean value V, and giving an estimated value of the target position by using a Kalman filter according to information of a target position and speed at the current moment; calculating the Mahalanobis distance between the predicted value of the target position at the previous moment and the measured value at the current moment, and when the calculated Mahalanobis distance is within the range of the set point track association threshold, considering that the point track is associated with the corresponding track; and predicting the trace points at each moment, and associating the related tracks to obtain a track tracking route of the target.

7. The millimeter wave radar-based 3D multi-target tracking processing method according to claim 1, wherein the step 5 specifically comprises: the trajectory states are divided into three types: detected, activated and deactivated; when the track is detected, counting the track, when the count is greater than an activation threshold, the track is in an activation state, otherwise, the track is invalid; when the track is in the activated state, counting the track, when the count is greater than a failure threshold value, the track is in the deactivated state, otherwise, the track is in the activated state.

Technical Field

The invention relates to the technical field of radar signal processing, in particular to a three-dimensional (3D) multi-target tracking processing method based on a millimeter wave radar, which can accurately detect and track multiple targets in a complex environment and determine the position of the target in the space from a three-dimensional space.

Background

The millimeter wave radar has the characteristics of interference resistance, instantaneity, all weather and the like, has the advantages of identifying a plurality of targets and the like, is more moderate in price compared with an expensive laser radar, cannot be influenced by external weather, and can detect information such as the distance and the speed of the targets. The method has rapid development in the civil field, and has wide application prospects in blind spot detection, traffic flow detection and altimeters.

The multi-target tracking of the millimeter wave radar mainly utilizes a plurality of data association strategy algorithms and Kalman filtering to estimate the state change of a moving object. Due to the diversity of the data association strategy algorithm, different tracking algorithms can be obtained. The method comprises the steps of obtaining data of a plurality of targets through the characteristics of a radar, forming a 3D point cloud on a target in space through clustering processing of point targets, obtaining a plurality of point cloud information according to the characteristics of the point targets, and dividing the point cloud based on target point density in a traditional DBSCAN algorithm. However, the method uses distance as a determination condition, accuracy of obtaining distance information at each stage is required, and multiple objects may meet each other during the movement process, and cannot be distinguished by the method.

With the rapid increase of urban traffic volume, in practical applications, for example, under complex conditions such as traffic flow statistics, in order to further promote intelligent traffic to replace manual monitoring, and in order to solve the above problems, it is of great significance to seek a multi-target tracking method.

Disclosure of Invention

Aiming at the problems in the prior art, the invention aims to provide a 3D multi-target tracking processing method based on a millimeter wave radar, which can realize multi-target tracking of radar signals and improve the reliability of multi-target detection.

The main technical thought of the invention is as follows: firstly, processing a millimeter wave radar baseband signal to obtain multiple-target point traces in space, selecting the point traces with strong signal-to-noise ratio to draw a 3D point cloud picture, and carrying out subsequent clustering and tracking processing according to the point cloud picture; the input of the clustering processing is a plurality of 3D point clouds with information in a spatial three-dimensional direction, the point traces which are adjacent to the searching distance position and are in the same category are traversed, the point traces in the same category are obtained through the marks in the same category, namely the result obtained through clustering, the parameters such as the spatial position of the point traces in the category are determined and then input into the track tracking, the point trace correlation and the point trace management of the track algorithm are carried out, the track of the point trace of each frame is obtained, the target is tracked, and the ID number of the target is marked.

In order to achieve the purpose, the invention is realized by adopting the following technical scheme.

A3D multi-target tracking processing method based on millimeter wave radar comprises the following steps:

step 1, a millimeter wave radar transmits electromagnetic waves and receives baseband echo signals, and the baseband echo signals are processed to obtain data information of a plurality of traces in space; the data information of each trace point comprises three-dimensional space position and speed information of the trace point;

step 2, clustering the plurality of traces through a clustering algorithm, and drawing the plurality of traces belonging to the same class into a 3D point cloud so as to obtain a plurality of target point clouds; removing noise point clouds from a plurality of target point clouds to obtain effective target point clouds;

step 3, determining an approximate space position and a speed average value V of a target according to the three-dimensional space position and speed information of the trace in the effective target point cloud; wherein the approximate spatial position of the target comprises a position of the target in the direction of the azimuthal X coordinate axisPosition of target in direction of orientation Y coordinate axisThe position of the target in the direction perpendicular to the Z coordinate axis is Z_max；

Step 4, according to the position of the target in the direction of the X coordinate axis of the azimuthPosition of target in direction of orientation Y coordinate axisCalculating to obtain a horizontal distance R and a horizontalAn angle A, determining a track tracking route of the target according to the horizontal distance R, the horizontal angle A and the speed mean value V;

and 5, carrying out track management on the track tracking route of the target, and updating the track state at any time.

The technical scheme of the invention has the characteristics and further improvements that:

(1) the step 2 specifically comprises the following substeps:

substep 2.1, the clustering algorithm firstly needs to perform traversal marking and category marking on each trace point, the default is 0, secondly needs to distinguish each trace point, and sets the spatial position information of the first point as (x)₀,y₀,z₀) Firstly, marking the point as detected, traversing the whole space by taking the point as a core point, and searching all similar points of the core point to obtain a plurality of points of the same type, which are called as point clouds of the target;

substep 2.2, setting the minimum clustering point number minPointsInCluster in one class, and if the number of the point traces in the same class is greater than the minimum clustering point number minPointsInCluster, determining the point traces as effective target point clouds; otherwise, it is noise point cloud.

(2) In sub-step 2.1, the method of finding all homogeneous points of the core point comprises the sub-steps of:

substep 2.1.1, determining core points (x)₀,y₀,z₀) Whether all points in the domain of (1) are homogeneous or not, the formula is:

wherein epsilon is the radius of the clustering algorithm field; v. of₀For the speed of the current core point, yFactor is a weighting factor for setting a Y coordinate, zFactor is a weighting factor for setting a Z coordinate, and vFactor is a weighting factor for setting a speed; (x, y, z) is three-dimensional spatial position information of any one point in the core point region, and v is velocity information of any one point in the core point region;

substep 2.1.2, statistics belong to core point (x)₀,y₀,z₀) The number of the point traces in the same class, numInEps, in the field if the number satisfies numInEps>Traversing and marking the trace points meeting the conditions under the minPointsInCluster condition;

and sequentially taking the traversably marked point traces as core points, and performing traversal search on the points which are not traversably marked until all the marked points are searched to obtain a plurality of points in the same category.

(3) In step 3, the position of the target in the direction of the azimuth X coordinate axisComprises the following steps:x_sumthe point clouds are all sums in the X direction of the point clouds in the same category, and N is the total number of the point traces in the category;

the position of the target in the direction of the orientation Y coordinate axisComprises the following steps:y_sumall sums in the Y direction of the point clouds in the same category;

the position of the target in the direction vertical to the Z coordinate axis is Z_maxThe maximum value of the point clouds in the same category in the direction vertical to the Z coordinate axis;

the speed mean value V is: v ═ V_sum/N，v_sumAll velocity sums for the same category of point clouds.

(4) In step 4, the horizontal distance R is:the horizontal angle A is as follows:

(5) in step 4, the determining the trajectory tracking route of the target according to the horizontal distance R, the horizontal angle a and the speed mean V specifically includes:

representing the moving state of the whole target by using a horizontal distance R, a horizontal angle A and a speed mean value V, and giving an estimated value of the target position by using a Kalman filter according to information of a target position and speed at the current moment; calculating the Mahalanobis distance between the predicted value of the target position at the previous moment and the measured value at the current moment, and when the calculated Mahalanobis distance is within the range of the set point track association threshold, considering that the point track is associated with the corresponding track; and predicting the trace points at each moment, and associating the related tracks to obtain a track tracking route of the target.

(6) The step 5 specifically comprises the following steps: the trajectory states are divided into three types: detected, activated and deactivated; when the track is detected, counting the track, when the count is greater than an activation threshold, the track is in an activation state, otherwise, the track is invalid; when the track is in the activated state, counting the track, when the count is greater than a failure threshold value, the track is in the deactivated state, otherwise, the track is in the activated state.

Compared with the prior art, the invention has the beneficial effects that:

the invention uses the millimeter wave radar with the detection characteristic of 3-dimensional space, not only can detect the distance, speed and azimuth angles from the radar to the target, but also can detect the pitch angle of the target, thereby drawing a multi-target 3D point cloud picture through clustering and the like and mastering the appearance contour of the target. The tracks of the multiple targets are obtained through a point track data association algorithm and Kalman filtering, and the calculated amount in the operation process can be greatly reduced through track management.

Drawings

The invention is described in further detail below with reference to the figures and specific embodiments.

FIG. 1 is an overall flow chart of a millimeter wave radar-based 3D multi-target tracking processing method of the present invention;

FIG. 2 is a diagram of a clustered secondary search domain;

FIG. 3 shows the effect of clustering three-dimensional space targets of single-target continuous multiframes;

FIG. 4 is a track diagram of a single-target continuous multi-frame target trace;

FIG. 5 is a track management flow diagram;

FIG. 6 is a diagram after two target continuous multiframe three-dimensional space point traces are clustered;

FIG. 7 is a trace diagram of the trace of the target point of two consecutive frames of targets.

Detailed Description

Embodiments of the present invention will be described in detail below with reference to examples, but it will be understood by those skilled in the art that the following examples are only illustrative of the present invention and should not be construed as limiting the scope of the present invention.

Referring to the flowchart of fig. 1, the millimeter-wave radar-based 3D multi-target tracking processing method of the present invention includes the following steps:

step 1, a millimeter wave radar transmits electromagnetic waves and receives baseband echo signals, and the baseband echo signals are processed to obtain data information of a plurality of traces in space; and the data information of each trace point comprises three-dimensional space position and speed information of the trace point.

Specifically, an echo obtained from an electromagnetic wave signal transmitted by the millimeter wave radar is subjected to baseband processing to obtain a plurality of traces containing three-dimensional spatial position and velocity information.

Step 2, clustering the plurality of traces through a clustering algorithm, and drawing the plurality of traces belonging to the same class into a 3D point cloud so as to obtain a plurality of target point clouds; and removing the noise point clouds from the plurality of target point clouds to obtain effective target point clouds.

Specifically, the clustering algorithm firstly needs to perform traversal marking and category marking on each trace point, the default is 0, and the purpose is to monitor each trace point. It should discriminate every trace point, assuming that the spatial information of the first point is (x)₀,y₀,z₀) It is marked as detected, and is used as a core point to traverse the whole space to find the homologous point. The judgment of the cluster needsThe spatial position (x, y, z), and velocity v are used to determine the domain of the core point of the target point cloud. Determining the radius epsilon of the clustering algorithm field according to the size of the target, setting the minimum clustering point number minPointsInCluster in one class, and recognizing the point trace number in one class as an effective target point only when the point trace number is more than the value (the minimum clustering point number minPointsInCluster), otherwise recognizing the point trace number as a noise point, thereby effectively avoiding the noise point from clustering into one class.

When a point is determined to be a core point, judging whether all points in the field are similar, wherein the formula is as follows:

wherein x is₀For the current core point at the azimuthal X coordinate position, y₀For the current core point at the azimuth Y-coordinate position, z₀For the current core point at the Z coordinate position, v₀For the speed of the current core point, yFactor is a weighting factor for setting a Y coordinate, zFactor is a weighting factor for setting a Z coordinate, and vFactor is a weighting factor for setting a speed.

The domain trace of the target core can be judged through the formula. For example, as shown in fig. 2, the core point is assumed to be point a, and the domains are found from the above expression as point c, point d, and point f. And when the number of the domain points numInEps and the minimum clustering point number minPointsInCluster meet the requirement that numInEps is greater than minPointsInCluster, the point c, the point d and the point f are considered to be the domain of the core point a. At this time, it cannot be considered that the point a finds all the domains, and a traversal search needs to be performed again to all the domains belonging to the core point a.

At this time, it is necessary to perform the domain search again on all the points in the searched domain, and as shown in fig. 2, the domain points c, d, and f searched in the previous step are traversed and marked. At this time, it is necessary to regard all the points traversed by the label as core points in turn, and then perform traversal search on the points not traversed by the label, and the search mode is determined as the formula above until the search of all the labeled points is completed. For example, regarding the point c as a core point, traversing all unmarked points to search the domain points b, h, m and n, and if the number of the domain points numInEps and the minimum clustering point number parameter minPointsInCluster do not satisfy numInEps > minPointsInCluster, regarding the point c as a boundary point of the category; on the contrary, if the expression is satisfied, the domain point b, the point h, the point m, the point n and the point c are considered as the same type of points. And similarly, obtaining the similar points of the category by the points d and f in the same searching mode, circularly searching and traversing according to the same local mode until the points of the category are searched, and stopping circulation. And after all the point searches are finished, obtaining a plurality of points of the same category, which are called the point clouds of the target. As shown in FIG. 3, the result of the three-dimensional space after an object is clustered, the shape of an object in a category can be roughly distinguished.

And 3, the point traces of the same type can approximately reflect the position and the contour of the target, and the space position and the speed state of each point trace are obtained through signal processing, so that the approximate space position, the speed state and the like of the target can be obtained. Taking the mean value of point clouds of the same category in the direction of an orientation X coordinate axis, and considering the mean value as the position of the target in the direction of the orientation X coordinate axis; taking the mean value of point clouds of the same category in the direction of an orientation Y coordinate axis, and considering the mean value as the position of the target in the direction of the orientation Y coordinate axis; taking the maximum value of the point clouds of the same category in the direction vertical to the Z coordinate axis, and considering the maximum value as the position of the target in the direction vertical to the Z coordinate axis; and taking the average speed of the point clouds in the same category, and considering the average speed as the speed of the target.

Specifically, an object shape and state and velocity may be determined by the trace of dots in the category. If all the sums of the X directions of the category point clouds are X_sumThen the position of the target in the direction of the azimuth X coordinate axis is considered asWherein N is the total number of traces in the category; if all the sums of the orientation Y coordinate axis directions of the point clouds in the category are Y_sumThen the position of the target in the direction of the azimuth Y coordinate axis is considered asIf the maximum value of the point cloud of the category in the direction vertical to the Z coordinate axis is Z_maxThen the position of the target in the direction perpendicular to the Z coordinate axis is considered as Z_max(ii) a If the sum of all the speeds of the point clouds in the category is v_sumThen the speed of the target is considered as V ═ V_sumN; the clustered target state can be obtained as described above.

Step 4, tracking the track: the spatial position of the target obtained by the last step of clustering is calculated by a formulaObtaining the horizontal distance R, the velocity V and the horizontal angle A, i.e.The moving state of the entire object is represented by this state. According to the information of a target position and speed at the current moment, the Kalman filter gives an estimated value of the target position, and due to the fact that the radar measurement interval is short, the vehicle is assumed to move at a constant speed at the next moment, and prediction is given to the target position at the next moment. The key step in achieving target trace tracking is data association and matching of traces. And calculating the mahalanobis distance between the predicted value of the target position at the previous moment and the measured value at the current moment, and when the calculated mahalanobis distance is within the set point-track association threshold range, considering that the point track is associated with the corresponding track. And predicting the trace points at each moment, and associating the related tracks to obtain a track tracking route of the target. As shown in fig. 4, the path is traced by a single target in a plurality of consecutive frames.

Step 5, track management: the track management is a key step for correctly associating the track, namely the state of updating the track at any moment, and the track state is divided into three types: detected, activated and deactivated. When the track is detected, counting the track, when the count is greater than an activation threshold, the track is in an activation state, otherwise, the track is invalid; when the track is in the activated state, counting the track, when the count is greater than a failure threshold value, the track is in the deactivated state, otherwise, the track is in the activated state. The state of the established track can be judged in real time through track management, and track association of next frame data only needs to consider whether to associate trace points in the effective track.

Specifically, as shown in fig. 5, the trajectory management of the target trajectory is a key step for correctly associating the trajectory, and the trajectory management is a state of updating the trajectory at any time, and the trajectory states are divided into three types: detected, activated and deactivated. When the track is in a detected state, namely state is 0, if the track is linked, namely flag is 1, the detection loss count is firstly set to be 0, when the detection activation count is greater than the activation threshold value, the track is in an activated state, otherwise, the detection activation count is increased by 1; otherwise, if the track is not linked to the trace point of the current frame, the track is detected to be lost and counted, when the detected lost count is larger than a failure threshold value, the track is marked as failure, otherwise, the detected lost count is increased by 1. When the track is in an active state, namely state is 1, if the track is linked, namely flag is 1, when the loss count is detected to be greater than 0, the loss count is detected to be reduced by 1; otherwise, if the track is not linked, the track is in a failure state when the detection loss count is greater than the failure threshold, otherwise the detection loss count is increased by 1. The state of the established track can be judged in real time through track management, and track association of next frame data only needs to consider whether the effective track and the track to be detected are associated with the trace points or not, and does not need to consider the track which is failed.

When a plurality of targets exist, the clustered effect of the two targets is obtained through the above steps, as shown in fig. 6. The tracks of multiple targets are distinguished through track association, for example, fig. 7 is a track diagram of two targets of continuous multiple frames, and ID numbers of different targets are respectively marked.

Although the present invention has been described in detail in this specification with reference to specific embodiments and illustrative embodiments, it will be apparent to those skilled in the art that modifications and improvements can be made thereto based on the present invention. Accordingly, such modifications and improvements are intended to be within the scope of the invention as claimed.