A millimeter wave radar point cloud clustering method

Abstract

The present invention relates to the application field of millimeter wave radar, and discloses a millimeter wave radar point cloud clustering method, which solves the problems of high complexity, low efficiency and low accuracy of millimeter wave radar point cloud heaping in the prior art. The scheme of the present invention first performs initial heaping and then performs heap merging processing based on the initially divided point cloud heap; in the initial heaping process, the first point is recorded as a heap, and then all the remaining points are traversed. If the current point is close to the previous point, it is classified as a heap, otherwise a new heap is created, and the heap information of the saved point heap is updated at the same time; after the traversal is completed, all points have corresponding heap numbers. In the heap merging process, all heaps are traversed to determine whether the heaps are adjacent to each other, and if they are adjacent, the heaps are merged.

Description

Technical Field

[0001] The present invention relates to the application field of millimeter wave radar, and in particular to a millimeter wave radar point cloud clustering method. Background Art

[0002] Millimeter-wave radar has all-weather characteristics, can accurately sense the tiny movements of targets, and does not form images, which can well protect the privacy of users. It is widely used in scenes such as personnel positioning, personnel detection, target tracking, and posture recognition in smart homes. The basis for the application of millimeter-wave radar depends on the clustering of point clouds. The quality of clustering has a great impact on the positioning, tracking, and posture of personnel.

[0003] The traditional method of dividing point clouds into piles is: record the first point as pile number 1, start traversing from the second point, calculate the distance between the two points, and if it is less than the proximity threshold, then record point 2 as pile number 1, then end the forward traversal of the current point, and start the forward traversal of the next point to determine the proximity. If the distance between the two points is greater than the proximity threshold, continue to traverse forward and determine until all previous points have no neighboring points, and the current point is recorded as a new pile point.

[0004] This method requires a lot of traversal times and has high time complexity when there are many scattered points. In addition, the point cloud is sorted by azimuth. If the distance difference of the point distribution generated by the target at a certain angle is quite different, then the point with the large difference will be given a new pile number, and the points at the adjacent angles will first be judged to be adjacent to the newly piled point. The subsequent points may be piled together with the new point pile. In this way, there may be an unreasonable situation where the points of a target are scattered at a certain angle, but the points at the adjacent angles are relatively concentrated, and the point will still be divided into two piles, and the accuracy of the pile division is low. Summary of the invention

[0005] The technical problem to be solved by the present invention is to propose a millimeter wave radar point cloud clustering method to solve the problems of high complexity, low efficiency and low accuracy of millimeter wave radar point cloud clustering in the prior art.

[0006] The technical solution adopted by the present invention to solve the above technical problems is:

[0007] A millimeter wave radar point cloud clustering method comprises the following steps:

[0008] S1. Process the received reflected millimeter-wave radar signal to obtain point cloud information;

[0009] S2. Calculate the three-dimensional coordinates of the point cloud according to the point cloud information, and perform coordinate rotation processing to obtain the reference system coordinates with the radar as the origin and the plane parallel to the ground as the XY plane;

[0010] S3, traverse the point cloud and perform initial heap division on each point:

[0011] Record the first point as pile number 1, and initialize the pile information of the corresponding pile number. Then traverse all the remaining points in the point cloud, compare the interval between each point and the previous point, and determine whether they are adjacent points. If so, record the pile number of the current point as the pile number of the previous point, and update the pile information of the corresponding pile number; otherwise, create a new pile number for the current point, the pile number is the pile number of the previous point + 1, and initialize the pile information of the new pile number;

[0012] S4, merging the point cloud piles after the initial grouping:

[0013] Sequentially traverse all the point cloud piles after the initial division, compare the pile information of the current point cloud pile with that of the other point cloud piles, and determine whether there is a point cloud pile adjacent to the current point cloud pile. If so, merge the pile information of the point cloud pile adjacent to the current point cloud pile into the pile information of the current point cloud pile. After completing the traversal of all the point cloud piles after the initial division, continue to traverse the point cloud piles after the merged processing until there are no point cloud piles that can be merged.

[0014] Furthermore, in step S1, the point cloud information includes: a distance range of the target relative to the radar, an azimuth angle α, a pitch angle θ, a signal-to-noise ratio SNR, and a Doppler speed doppler.

[0015] Furthermore, in step S2, the three-dimensional coordinates of the point cloud are calculated according to the point cloud information, specifically including:

[0016] The x coordinate of the point is obtained by the formula x=range*cos(θ)*sin(α);

[0017] The y coordinate of the point is obtained by the formula y = range*cos(θ)*cos(α);

[0018] The z coordinate of the point is obtained by the formula z=range*sin(θ).

[0019] Further, in step S2, the coordinate rotation process is performed to obtain a reference system coordinate with the radar as the origin and a plane parallel to the ground as the XY plane, specifically including:

[0020] x' = x

[0021] y'=cosα·y-sinα·z

[0022] z'=sinα·y+sinα·z

[0023] Wherein, x, y, z are the x, y, z coordinates of the point before the coordinate rotation processing; x', y', z' are the x, y, z coordinates after the coordinate rotation processing.

[0024] Further, in step S3, the pile information includes the distribution range of the point cloud pile, that is, the maximum x coordinate maxX, the maximum y coordinate maxY, the maximum z coordinate maxZ, the minimum x coordinate minX, the minimum y coordinate minY and the minimum z coordinate minZ of all points in the point cloud pile.

[0025] Further, in step S3, the comparing the interval between each point and the previous point to determine whether they are adjacent points specifically includes: setting spaceX, spaceY, and spaceZ as the x interval, y interval, and z interval between the two points, and when spaceX, spaceY, and spaceZ simultaneously meet the threshold, the two points are determined to be adjacent points; wherein the x interval, y interval, and z interval are the intervals between the two points in the x-axis, y-axis, and z-axis directions, respectively.

[0026] Furthermore, in step S4, comparing the pile information of the current point cloud pile with that of other point cloud piles to determine whether there is a point cloud pile adjacent to the current point cloud pile specifically includes:

[0027] Determine whether two point cloud piles meet the non-adjacent condition. If so, the two point cloud piles are non-adjacent. Otherwise, the two point cloud piles are determined to be adjacent.

[0028] Among them, the non-adjacent conditions include:

[0029] For two point cloud piles A and B, we have:

[0030] A.maxY+spaceY<B.minY||A.minY-spaceY> B.maxY||A.maxX+spaceX <B.minx||

[0031] A.minX-spaceX>B.maxX||A.maxZ+spaceZ<B.minZ||A.minZ-spaceZ> B.maxZ

[0032] Among them, A.maxX, A.minX, A.maxY, A.minY, A.maxZ, and A.minZ represent the maximum x coordinate, minimum x coordinate, maximum y coordinate, minimum y coordinate, maximum z coordinate, and minimum z coordinate in the point cloud pile A, respectively;

[0033] B.maxX, B.minX, B.maxY, B.minY, B.maxZ, and B.minZ represent the maximum x coordinate, minimum x coordinate, maximum y coordinate, minimum y coordinate, maximum z coordinate, and minimum z coordinate in the point cloud pile B, respectively;

[0034] spaceX, spaceY, and spaceZ represent the intervals between point cloud pile A and point cloud pile B in the x-axis, y-axis, and z-axis directions, respectively.

[0035] The beneficial effects of the present invention are:

[0036] (1) The scheme of the present invention first determines the adjacent points according to the interval between the points in the point cloud, and divides the adjacent points into the same point cloud pile, thereby performing initial heap division; then the point cloud pile is traversed, and the adjacent point cloud piles are combined into a pile. The point cloud clustering and heap division using this scheme will not result in adjacent points being divided into two piles, and the heap division is more reasonable and more accurate;

[0037] (2) In the initial heap division process, the solution of the present invention only compares the interval between the current point and the previous point to determine whether they are adjacent points. The adjacent points are divided into the same point cloud heap as the previous point. If they are non-adjacent points, a new heap is directly created to complete the initial heap division. This process does not require excessive traversal, the heap division complexity is low, and the efficiency is higher.

[0038] (3) When initially dividing the stack to determine whether two points are adjacent points, the solution of the present invention does not directly calculate the distance between the two points, but directly calculates the interval between the two points in the x, y, and z axes, without the need to process square and square root operations, thus reducing the computational workload; in the process of stacking the point cloud, the method of excluding non-adjacent stacks is used to determine whether the point cloud stack is an adjacent stack, without considering the complex conditions for determining adjacent stacks, thus reducing the computational complexity. Therefore, the efficiency of stacking can be further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of the millimeter wave radar point cloud clustering method in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The present invention aims to propose a millimeter wave radar point cloud clustering method to solve the problems of high complexity, low efficiency and low accuracy of millimeter wave radar point cloud clustering in the prior art. The solution of the present invention first performs initial clustering and then performs clustering processing based on the initially divided point cloud clusters; in the initial clustering process, the first point is recorded as a cluster, and then all the remaining points are traversed. If the current point is close to the previous point, it is classified as a cluster, otherwise a new cluster is created, and the cluster information of the saved point cluster is updated at the same time; after the traversal is completed, all points have corresponding cluster numbers. In the clustering process, all clusters are traversed to determine whether the clusters are adjacent to each other, and if they are adjacent, they are clustered.

[0041] Example:

[0042] like Figure 1 As shown, the millimeter wave radar point cloud clustering method in this embodiment includes the following steps:

[0043] 1. Process the received reflected millimeter-wave radar signal to obtain point cloud information:

[0044] Due to the radar antenna array, the upper and lower visual range is relatively narrow, so the radar needs to be tilted downward at a certain angle when installed to minimize the blind area below the radar. The radar sends electromagnetic wave signals to the space to be measured, and the received reflected signals are processed to obtain dynamic point cloud information data in the space to be measured, which includes the distance range of the target relative to the radar, azimuth angle α, pitch angle θ, signal-to-noise ratio SNR, Doppler speed doppler and other information.

[0045] 2. Calculate the three-dimensional coordinates of the point cloud based on the point cloud information, and perform coordinate rotation processing to obtain the reference system coordinates with the radar as the origin and the plane parallel to the ground as the XY plane;

[0046] In this step, the three-dimensional coordinates of the point cloud are first calculated based on the point cloud information. Specifically, the X coordinate can be obtained by the formula range*cos(θ)*sin(α), the Y coordinate can be obtained by the formula range*cos(θ)*cos(α), and the Z coordinate can be obtained by range*sin(θ).

[0047] The three-dimensional coordinates calculated here are all reference system coordinates with the radar as the origin, the radar antenna plane as the XZ plane, and the direction perpendicular to the radar antenna plane forward as the positive direction of the Y axis. Since the radar is installed at an angle, the radar area detection effect is better, and its tilt angle is equivalent to a certain angle rotation around the X axis. Then it is also necessary to perform coordinate rotation processing on the three-dimensional coordinates calculated above, so as to convert them into reference system coordinates with the radar as the origin and the plane parallel to the ground as the XY plane. Specifically:

[0048] x' = x

[0049] y'=cosα·y-sinα·z

[0050] z'=sinα·y+sinα·z

[0051] Wherein, x, y, z are the x, y, z coordinates of the point before the coordinate rotation processing; x', y', z' are the x, y, z coordinates after the coordinate rotation processing.

[0052] 3. Traverse the point cloud and perform initial heap division on each point:

[0053] The first point cloud is recorded as pile number 1, and the pile information of the corresponding pile number is initialized. The pile information must at least contain the distribution range of the point pile, that is, maxX, maxY, maxZ, minX, minY, minZ, that is, the point cloud divided into a pile is wrapped with a minimum rectangular body. When there is only one point, the maximum and minimum are the same value. Then start traversing from the second point, and compare the distance between each point and the previous point. If the distance is less than the set threshold, the current point and the previous point are close points, and the pile number of the current point is recorded as the pile number of the previous point, and the point pile information is updated. If the distance is greater than the set threshold, it will no longer continue to traverse forward like the traditional method, but will create a new pile number, which is the pile number of the previous point plus 1, and initialize the pile information of the pile at the same time. Since the distance formula involves squares and square roots, and embedded computing resources are limited, in order to reduce the amount of calculation, it is not necessary to directly calculate the distance between two points. SpaceX, spaceY, and spaceZ can be set as the x interval, y interval, and z interval between two points. Only when spaceX, spaceY, and spaceZ meet the threshold at the same time are they close points. After the traversal is completed, each point has a corresponding heap number. The advantage of this is that you only need to traverse once and simply divide the heap. For example, if there are 100 points, you only need to compare 99 times to complete the initial heap division.

[0054] 4. Combine the point cloud piles after the initial separation:

[0055] In this step, all the piles are traversed sequentially, and the distribution information of the current pile is compared with the distribution information of the remaining piles to see if there is overlap or proximity. If so, the pile information is merged and classified into one pile. After updating the pile information, the traversal continues until there is no pile to be merged, and the heap clustering is completed. Here, to determine whether the boundaries of the point piles overlap or are adjacent, the maximum interval in the x direction can be set to spaceX, the maximum interval in the y direction to spaceY, and the maximum interval in the z direction to spaceZ. As long as the intervals in all directions between the point piles are less than the set threshold, they are adjacent point piles. Since there are many forms of overlap or proximity, it is impossible to judge each one, and the conditions are too complicated, so only non-adjacent scenarios need to be judged. For example, if there are two piles of points A and B, the non-adjacent conditions can be set to:

[0056] A.maxY+spaceY<B.minY||A.minY-spaceY> B.maxY||A.maxX+spaceX <B.minx||

[0057] A.minX-spaceX>B.maxX||A.maxZ+spaceZ<B.minZ||A.minZ-spaceZ> B.maxZ

[0058] Among them, A.maxX, A.minX, A.maxY, A.minY, A.maxZ, and A.minZ represent the maximum x coordinate, minimum x coordinate, maximum y coordinate, minimum y coordinate, maximum z coordinate, and minimum z coordinate in the point cloud pile A, respectively;

[0059] B.maxX, B.minX, B.maxY, B.minY, B.maxZ, and B.minZ represent the maximum x coordinate, minimum x coordinate, maximum y coordinate, minimum y coordinate, maximum z coordinate, and minimum z coordinate in the point cloud pile B, respectively;

[0060] spaceX, spaceY, and spaceZ represent the intervals between point cloud pile A and point cloud pile B in the x-axis, y-axis, and z-axis directions, respectively.

[0061] In practical applications, the proximity threshold can be flexibly set according to different functional requirements. For example, in posture judgment, the point cloud generated by a person when falling is relatively scattered. The proximity threshold can be appropriately raised to try to group the scattered points together to improve the accuracy of the posture.

[0062] Finally, it should be noted that the above embodiments are only preferred implementations and are not intended to limit the present invention. It should be pointed out that for those skilled in the art, several modifications, equivalent replacements, improvements, etc. can be made without departing from the scope of the present invention and the scope of protection of the claims, and all of these should be included in the protection scope of the present invention.

Claims

1. A millimeter wave radar point cloud clustering method, characterized in that: The following steps are involved: S1. Process the received reflected millimeter-wave radar signal to obtain point cloud information; S2. Calculate the three-dimensional coordinates of the point cloud according to the point cloud information, and perform coordinate rotation processing to obtain the reference system coordinates with the radar as the origin and the plane parallel to the ground as the XY plane; S3, traverse the point cloud and perform initial heap division on each point: Record the first point as pile number 1, and initialize the pile information of the corresponding pile number. Then traverse all the remaining points in the point cloud, compare the interval between each point and the previous point, and determine whether they are adjacent points. If so, record the pile number of the current point as the pile number of the previous point, and update the pile information of the corresponding pile number; otherwise, create a new pile number for the current point, the pile number is the pile number of the previous point + 1, and initialize the pile information of the new pile number; S4, merging the point cloud piles after the initial grouping: Sequentially traverse all the point cloud piles after the initial division, compare the pile information of the current point cloud pile with that of the other point cloud piles, and determine whether there is a point cloud pile adjacent to the current point cloud pile. If so, merge the pile information of the point cloud pile adjacent to the current point cloud pile into the pile information of the current point cloud pile. After completing the traversal of all the point cloud piles after the initial division, continue to traverse the point cloud piles after the merged processing until there are no point cloud piles that can be merged.

2. A millimeter wave radar point cloud clustering method as claimed in claim 1, characterized in that: In step S1, the point cloud information includes: the distance range of the target relative to the radar, the azimuth angle α, the pitch angle θ, the signal-to-noise ratio SNR and the Doppler speed doppler.

3. A millimeter wave radar point cloud clustering method as claimed in claim 2, characterized in that: In step S2, the three-dimensional coordinates of the point cloud are calculated according to the point cloud information, specifically including: The x coordinate of the point is obtained by the formula x=range*cos(θ)*sin(α); The y coordinate of the point is obtained by the formula y = range*cos(θ)*cos(α); The z coordinate of the point is obtained by the formula z=range*sin(θ).

4. A millimeter wave radar point cloud clustering method as claimed in claim 2, characterized in that: In step S2, the coordinate rotation process is performed to obtain the reference coordinate system with the radar as the origin and the plane parallel to the ground as the XY plane, which specifically includes: x' = x y'=cosα·y-sinα·z z'=sinα·y+sinα·z Wherein, x, y, z are the x, y, z coordinates of the point before the coordinate rotation processing; x', y', z' are the x, y, z coordinates after the coordinate rotation processing.

5. The millimeter wave radar point cloud clustering method according to claim 1, characterized in that: In step S3, the pile information includes the distribution range of the point cloud pile, that is, the maximum x coordinate maxX, the maximum y coordinate maxY, the maximum z coordinate maxZ, the minimum x coordinate minX, the minimum y coordinate minY and the minimum z coordinate minZ of all points in the point cloud pile.

6. A millimeter wave radar point cloud clustering method as claimed in claim 1, characterized in that: In step S3, the comparison of the interval between each point and the previous point to determine whether they are adjacent points specifically includes: setting spaceX, spaceY, and spaceZ as the x interval, y interval, and z interval between the two points, and when spaceX, spaceY, and spaceZ meet the threshold at the same time, determining that the two points are adjacent points; wherein the x interval, y interval, and z interval are the intervals between the two points in the x-axis, y-axis, and z-axis directions, respectively.

7. A millimeter wave radar point cloud clustering method according to any one of claims 1 to 6, characterized in that: In step S4, comparing the pile information of the current point cloud pile with that of other point cloud piles to determine whether there is a point cloud pile adjacent to the current point cloud pile specifically includes: Determine whether two point cloud piles meet the non-adjacent condition. If so, the two point cloud piles are non-adjacent. Otherwise, the two point cloud piles are determined to be adjacent. Among them, the non-adjacent conditions include: For two point cloud piles A and B, we have: A.maxY+spaceY<B.minY||A.minY-spaceY> B.maxY||A.maxX+spaceX<B.minx||A.minX-spaceX> B.maxX||A.maxZ+spaceZ<B.minZ||A.minZ-spaceZ> B.maxZ Among them, A.maxX, A.minX, A.maxY, A.minY, A.maxZ, and A.minZ represent the maximum x coordinate, minimum x coordinate, maximum y coordinate, minimum y coordinate, maximum z coordinate, and minimum z coordinate in the point cloud pile A, respectively; B.maxX, B.minX, B.maxY, B.minY, B.maxZ, and B.minZ represent the maximum x coordinate, minimum x coordinate, maximum y coordinate, minimum y coordinate, maximum z coordinate, and minimum z coordinate in the point cloud pile B, respectively; spaceX, spaceY, and spaceZ represent the intervals between point cloud pile A and point cloud pile B in the x-axis, y-axis, and z-axis directions, respectively.

Images