A method for detecting dynamic targets in an indoor environment based on a vehicle-mounted sensor system

Abstract

The application discloses a kind of indoor environment dynamic target detection methods based on vehicle-mounted sensor system, collect laser radar data and millimeter wave radar data;Using DBSCAN algorithm, the laser radar data after pre-processing is pre-clustered, according to each direction point cloud distribution variance mean, update Euclidean distance calculation formula, using the updated Euclidean distance calculation formula, the laser radar data after pre-processing is improved DBSCAN clustering, obtain laser radar target final clustering cluster set;The position of laser radar target is tracked using the extended Kalman filter tracker, the displacement, speed and motion direction information of target are calculated;Doppler velocity and azimuth information are extracted from millimeter wave radar data, the mapping of Doppler velocity in vehicle motion direction is calculated, the vehicle motion direction speed value distribution curve is calculated, and the speed value at the peak of distribution curve is the vehicle self-motion speed;Calculate target absolute motion speed, judge whether it is moving target.The application can reduce algorithm overhead, ensure detection real-time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traffic target state detection, and in particular to an indoor environment dynamic target detection method based on a vehicle-mounted sensor system. BACKGROUND

[0002] Moving obstacles are uncertain factors in the scene and are the key perception objects in the indoor intelligent driving link. In the SLAM mapping process of automatic driving, the existence of dynamic targets will lead to inaccurate mapping. In view of the poor mapping stability of the existing vehicle-mounted sensor system in the dynamic indoor scene, the present application proposes an indoor environment dynamic target detection method based on a vehicle-mounted sensor system.

[0003] At present, the research on dynamic target detection based on the vehicle-mounted sensor system includes the patent "Laser SLAM method fusing dynamic target estimation". In the invention, a point cloud target detection network PointRCNN is used to perform three-dimensional target detection on point cloud data, a laser odometer based on the nearest point iteration method is used to estimate the motion of the unmanned vehicle, and a target state estimation method is designed to divide the targets in the scene into static and dynamic.

[0004] The application scenario of the method is outdoor. In the outdoor scene, there are many feature points of laser radar data, and in the indoor scene, there are high-repetition scenes such as corridors. If the indoor scene is processed by the method, the laser odometer based on the nearest point iteration method cannot perform accurate motion estimation, and at the same time, the nearest point iteration algorithm has a huge amount of calculation and cannot meet the real-time requirement.

[0005] In view of the problems existing in the existing algorithm, the present application builds an algorithm framework for obtaining the relative motion state of the target by using the laser radar for target tracking and using the millimeter wave radar for vehicle self-motion estimation. The algorithm can well adapt to the high-similarity indoor scene, realize accurate motion target detection, and the millimeter wave radar self-motion compensation algorithm has a very small performance overhead to ensure the real-time performance of the system. At the same time, the algorithm in the present application is adapted to low-performance hardware, and accurate target relative position and speed information can be obtained by using a low-beam laser radar. SUMMARY

[0006] The purpose of the present application is to provide an indoor environment dynamic target detection method based on a vehicle-mounted sensor system in view of the problems existing in the above-mentioned existing indoor environment dynamic target detection.

[0007] The technical solution for achieving the purpose of the present application is: an indoor environment dynamic target detection method based on a vehicle-mounted sensor system, comprising the following steps:

[0008] Step 1, data acquisition is performed by using a vehicle-mounted sensor system to obtain a data packet, including laser radar data and millimeter wave radar data;

[0009] Step 2, extract the lidar data in the data packet, and pre-process the lidar data, including ROI region extraction and point cloud downsampling;

[0010] Step 3, using the DBSCAN algorithm to pre-cluster the pre-processed lidar data, calculating the mean of the point cloud distribution variance of all clustering clusters in each direction in the pre-clustering result, and updating the Euclidean distance calculation formula according to the mean of the point cloud distribution variance in each direction, using the updated Euclidean distance calculation formula to perform improved DBSCAN clustering on the pre-processed lidar data, and obtaining a final clustering cluster set of lidar targets;

[0011] Step 4, calculating the centroid of each clustering cluster in the final clustering cluster set of lidar targets as the position of the lidar target corresponding to the clustering cluster, constructing an extended Kalman filter tracker to track the position of the lidar target, and calculating the displacement, speed and motion direction information of the target according to the target moving trajectory obtained by tracking;

[0012] Step 5, extracting the millimeter wave radar data in the data packet, extracting the Doppler velocity and azimuth angle information from the millimeter wave radar data, calculating the mapping of the Doppler velocity in the vehicle motion direction, obtaining the vehicle motion direction speed value, calculating the vehicle motion direction speed value distribution curve, finding the speed value at the peak of the distribution curve, and identifying this speed value as the vehicle self-motion speed;

[0013] Step 6, fusing the target speed and motion direction information obtained by the extended Kalman filter tracker and the vehicle self-motion speed, calculating the target absolute motion speed, and judging whether it is a moving target.

[0014] Further, in step 2, the lidar data in the data packet is extracted, and the lidar data is pre-processed, including ROI region extraction and point cloud downsampling, and the specific method is:

[0015] Extract the lidar data in the data packet, set the ROI region extraction range threshold x min , x max , y min , y max , z min , z max according to the effective detection range of the low-beam lidar in the indoor scene;

[0016] Using the ROI region extraction range threshold to judge each point cloud in the lidar data, retaining the points within the ROI extraction range threshold, and filtering out the points within the extraction range threshold, to obtain the lidar data within the ROI range;

[0017] The laser radar data in the ROI range is down-sampled using a voxel-based down-sampling method to obtain pre-processed laser radar data.

[0018] Further, in step 3, the pre-processed laser radar data is pre-clustered using a DBSCAN algorithm, the mean of the point cloud distribution variance of all cluster groups in each direction is calculated, and the Euclidean distance calculation formula is updated according to the mean of the point cloud distribution variance in each direction, the improved DBSCAN clustering of the pre-processed laser radar data is performed using the updated Euclidean distance calculation formula, and the final cluster group set of the laser radar target is obtained, and the specific method is as follows:

[0019] Step 3.1, preset the pre-clustering parameters neighborhood radius eps' and minimum point number Minpts', pre-cluster the pre-processed laser radar data using the DBSCAN algorithm, and the pre-clustering result is represented as {M'1, M'2, …, M' n}, M' n = {(x'1, y'1, z'1), (x'2, y'2, z'2), …, (x' m , y' m , z' m )} is a cluster group in the pre-clustering result, which contains a plurality of laser radar points; the mean of the x-axis direction point cloud distribution variance of all cluster groups is calculated based on the coordinates of all points in each cluster group M' n the mean of the y-axis direction point cloud distribution variance the mean of the z-axis direction point cloud distribution variance The calculation process is described as follows:

[0020]

[0021]

[0022]

[0023] Step 3.2, according to the mean of the point cloud distribution variance in each direction update the weight parameters α, β, δ of the x, y, z axis direction distance in the Euclidean distance calculation formula to obtain the improved Euclidean distance calculation formula, and the specific design of the improved Euclidean distance calculation formula and the parameter calculation process are as follows:

[0024]

[0025] wherein,

[0026]

[0027] ​Step 3.3, set improved DBSCAN clustering parameters neighborhood radius eps and minimum point number Minpts, in the improved DBSCAN clustering process, use the improved Euclidean distance calculation formula to calculate the distance r between points in the preprocessed laser radar data, two points meeting r < eps belong to the same temporary clustering cluster, and the set {M1, M2, …, M n} composed of temporary clustering clusters meeting the point number greater than Minpts is the final clustering cluster set of the improved DBSCAN clustering algorithm. n m m m .

[0028] Further, step 4, calculate the centroid of each clustering cluster in the final clustering cluster set of the laser radar target as the position of the laser radar target corresponding to the clustering cluster, construct an extended Kalman filter tracker to track the position of the laser radar target, and calculate the speed and motion direction information of the target according to the target moving track obtained by tracking, and the specific method is:

[0029] Step 4.1, traverse each clustering cluster M n in the final clustering cluster set {M1, M2, …, M n} of the laser radar target, and calculate the centroid (x, y, z) of the point set in the clustering cluster M n by taking the mean value of the coordinates of each point in the clustering cluster M n in the x, y, z axis direction as the corresponding target position, and calculate the three-dimensional size information (l, w, h) of the target by taking the maximum coordinates x max , y max , z max and the minimum coordinates x min , y min , z min of the clustering cluster M n in the x, y, z axis direction, and the specific calculation process is as follows:

[0030]

[0031] Step 4.2, splice the target position and the three-dimensional size information of the target to obtain the target state matrix (x, y, z, l, w, h) at the current time. Data association is performed between the target state matrix at the current time and the predicted value of the existing track, and the centroid displacement ratio and dimension similarity are taken as the basis for data association. If the predicted value of the existing track and the centroid displacement ratio of the target at the current time are less than the threshold value and the dimension similarity is less than the threshold value, it is considered that the two match, otherwise it is considered not to match; the state matrix of the predicted value of the existing track is (x A , y A , z​A l A ,w A ,h A ), the centroid displacement ratio calculation method is:

[0032]

[0033] The dimension similarity calculation method is as follows:

[0034]

[0035] Wherein, x_dim is the x dimension similarity, y_dim is the y dimension similarity, and z_dim is the z dimension similarity.

[0036] According to the matching result, the corresponding operation is performed:

[0037] If the current detected pedestrian target and the existing trajectory match successfully, the trajectory is updated based on the extended Kalman filter update function using the current target state matrix;

[0038] If the current detected pedestrian target is not matched successfully, a new tracking trajectory is initialized with (x, y, z, l, w, h) as the starting position;

[0039] If the existing trajectory is not successfully matched, the coast_cycles of the lost association is increased by one, and compared with the threshold value, if greater than the threshold value, the trajectory is deleted;

[0040] According to the target state matrix in the matching result, the tracking trajectory list is updated, and the moving trajectory of the pedestrian target is output, then the two-dimensional coordinates of the current detected target are (x, y), and the coordinates of the target corresponding to the previous N frames of the trajectory are (x -N ,y -N ), the displacement calculation method of the current detected target in N frames is:

[0041]

[0042] The speed calculation method of the current detected target in N frames is:

[0043]

[0044] Wherein, Δt is the time difference, and the motion direction calculation method of the current detected target is:

[0045]

[0046] Further, step 5, extract the millimeter wave radar data in the data packet, extract the Doppler velocity and azimuth angle information from the millimeter wave radar data, calculate the mapping of the Doppler velocity in the vehicle motion direction, obtain the vehicle motion direction speed value, calculate the vehicle motion direction speed value distribution curve, find the speed value at the peak of the distribution curve, and determine that the speed value is the vehicle self-motion speed. The specific method is:

[0047] Extract the Doppler velocity and azimuth angle information in the millimeter wave radar data in the data packet. Calculate the speed of each point in the vehicle motion direction using the radial velocity and azimuth angle:

[0048]

[0049] Where, v 车 is the vehicle motion direction speed, v 多普勒 is the Doppler velocity in the millimeter wave radar data, and β is the azimuth angle in the millimeter wave radar data.

[0050] Sort v 车 , and use the Gaussian kernel function to perform kernel density estimation of v 车 to obtain the probability density function of the vehicle driving direction speed v 车 . The v 车 corresponding to the peak position of the probability density function is the vehicle self-motion speed v ego .

[0051] Further, step 6, fuse the target speed and motion direction information obtained by the extended Kalman filter tracker and the vehicle self-motion speed to calculate the target absolute motion speed and judge whether it is a moving target. The specific steps are:

[0052] Calculate the x-axis relative speed:

[0053]

[0054] Calculate the y-axis relative speed:

[0055]

[0056] Calculate the x-axis absolute speed:

[0057]

[0058] Calculate the y-axis relative speed:

[0059]

[0060] According to the speed, judge whether the target is a dynamic target:

[0061]

[0062] The application discloses an indoor environment dynamic target detection system based on a vehicle-mounted sensor system, and implements the indoor environment dynamic target detection method based on the vehicle-mounted sensor system, and realizes the indoor environment dynamic target detection based on the vehicle-mounted sensor system.

[0063] Compared with the prior art, the application has the following advantages: the millimeter wave radar is used as a mileage counter to estimate the self-motion of a vehicle, can well adapt to an indoor high-repetition scene, and greatly reduces the algorithm overhead compared with a feature point matching algorithm of a laser radar mileage counter, thereby guaranteeing the real-time performance of the algorithm. Meanwhile, the application improves the DBSCAN algorithm, improves the performance of the dynamic target detection algorithm, and can well adapt to a low-performance laser radar. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 It is a whole scheme diagram of the indoor environment dynamic target detection method based on the vehicle-mounted sensor system.

[0065] Figure 2 It is a specific flowchart of the indoor environment dynamic target detection method based on the vehicle-mounted sensor system.

[0066] Figure 3 It is a self-motion estimation principle of the millimeter wave radar.

[0067] Figure 4 It is a self-motion estimation result diagram of the millimeter wave radar.

[0068] Figure 5 It is an indoor environment dynamic target detection result diagram. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical scheme and advantages of the application clearer, the application is further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application and not used to limit the application.

[0070] The application filters and pre-processes original point cloud data collected by the laser radar to reduce the data amount, proposes an improved DBSCAN algorithm based on the anisotropy of the point cloud sparsity, uses the improved DBSCAN algorithm to cluster the pre-processed point cloud, uses the extended Kalman filter and the Hungarian matching to track the target, calculates the displacement, speed and motion direction of the target, constructs a millimeter wave radar mileage counter to estimate the self-motion state of the unmanned vehicle, uses a low-computing amount method without feature point matching to compensate the self-motion of the unmanned vehicle, obtains the motion state change information of the target, and performs dynamic target detection.

[0071] The indoor environment dynamic target detection method based on the vehicle-mounted sensor system comprises the following steps:

[0072] Step 1, data acquisition using vehicle-mounted sensor system to obtain data packets, including lidar data and millimeter wave radar data.

[0073] Step 2, extract lidar data in the data packet, pre-process the lidar data, including ROI region extraction and point cloud downsampling, the specific method is:

[0074] Extract the lidar data in the data packet, set the ROI region extraction range threshold x according to the effective detection range of the low-beam lidar in the indoor scene min , x max , y min , y max , z min , z max ;

[0075] Use the ROI region extraction range threshold to judge each point cloud in the lidar data, retain the points within the ROI extraction range threshold, filter out the points within the extraction range threshold, and obtain the lidar data within the ROI range;

[0076] Use the voxel-based downsampling method to downsample the lidar data within the ROI range to obtain the pre-processed lidar data.

[0077] Step 3, use DBSCAN algorithm to pre-cluster the pre-processed lidar data, calculate the mean of point cloud distribution variance of all cluster clusters in each direction in the pre-clustering result, and update the Euclidean distance calculation formula according to the mean of point cloud distribution variance in each direction, use the updated Euclidean distance calculation formula to perform improved DBSCAN clustering on the pre-processed lidar data, and obtain the final cluster set of lidar targets, the specific method is:

[0078] Step 3.1, pre-set the pre-clustering parameters neighborhood radius eps' and minimum point number Minpts', use the DBSCAN algorithm to pre-cluster the pre-processed lidar data, the pre-clustering result is represented as {M'1,M'2,…,M' n}, M' n = {(x'1,y'1,z'1),(x'2,y'2,z'2),…,(x' m ,y' m ,z' m )} is a cluster cluster in the pre-clustering result, which contains multiple lidar points; calculate the mean of x-axis direction point cloud distribution variance of all cluster clusters based on the coordinates of all points in each cluster M' n y-axis direction point cloud distribution variance mean z-axis direction point cloud distribution variance mean​ The calculation process is described as follows:

[0079]

[0080]

[0081]

[0082] Step 3.2, according to the mean of the variance of the point cloud distribution in each direction Update the weight parameters a, b, and d of the x, y, and z axis direction distance in the Euclidean distance calculation formula, obtain the improved Euclidean distance calculation formula, and the specific design of the improved Euclidean distance calculation formula and the parameter calculation process are as follows:

[0083]

[0084] Wherein,

[0085]

[0086] Step 3.3, set the improved DBSCAN clustering parameters neighborhood radius eps and minimum point number Minpts, in the improved DBSCAN clustering process, use the improved Euclidean distance calculation formula to calculate the distance r between the points in the preprocessed laser radar data, the two points satisfying r < eps belong to the same temporary clustering cluster, and the temporary clustering cluster satisfying the point number greater than Minpts forms a set {M1, M2, …, M n}, that is, the final clustering cluster set of the improved DBSCAN clustering algorithm, wherein M n = {(x1, y1, z1), (x2, y2, z2), …, (x m , y m , z m}.

[0087] Step 4, calculate the centroid of each clustering cluster in the final clustering cluster set of the laser radar target as the position of the laser radar target corresponding to the clustering cluster, construct an extended Kalman filter tracker to track the position of the laser radar target, and calculate the speed and motion direction information of the target according to the target moving track obtained by tracking. The specific method is:

[0088] Step 4.1, traverse each clustering cluster M n in the final clustering cluster set {M1, M2, …, M n}, calculate the centroid (x, y, z) of the point set in the clustering cluster M n by calculating the mean of the x, y, and z axis direction coordinates of each point in the clustering cluster M nThe maximum coordinates x, y, z and the minimum coordinates x, y, z in the x, y, z axis direction max max max min min min The target three-dimensional size information (l, w, h) is calculated as follows:

[0089]

[0090] Step 4.2, the target position is spliced with the target three-dimensional size information to obtain the current time target state matrix (x, y, z, l, w, h). The current time target state matrix is data associated with the predicted value of the existing trajectory, and the centroid displacement ratio and dimension similarity are used as the basis for data association. If the predicted value of the existing trajectory and the centroid displacement ratio of the current time target are less than the threshold value and the dimension similarity is less than the threshold value, it is considered that the two match, otherwise it is considered not to match; the state matrix of the existing trajectory predicted value is (x A , y A , z A , l A , w A , h A ), and the centroid displacement ratio is calculated as follows:

[0091]

[0092] The dimension similarity is calculated as follows:

[0093]

[0094] Wherein, x_dim is the x dimension similarity, y_dim is the y dimension similarity, and z_dim is the z dimension similarity.

[0095] According to the matching result, the corresponding operation is performed:

[0096] If the current detected pedestrian target and the existing trajectory match successfully, the extended Kalman filter update function is used to update the trajectory based on the current time target state matrix;

[0097] If the current detected pedestrian target is not matched successfully, a new tracking trajectory is initialized with (x, y, z, l, w, h) as the starting position;

[0098] If the existing trajectory is not successfully matched, the coast_cycles of the lost association is increased by one, and compared with the threshold value, if greater than the threshold value, the trajectory is deleted;

[0099] ​​​​​According to the matching result, the target state matrix at the current time is updated, the moving track of the pedestrian target is output, the two-dimensional coordinates of the current detection target are (x, y), the coordinates of the target corresponding to the previous N frames are (x -N ,y -N ), and the displacement of the current detection target in N frames is calculated as follows:

[0100]

[0101] The speed of the current detection target in N frames is calculated as follows:

[0102]

[0103] The motion direction of the current detection target is calculated as follows:

[0104]

[0105] Step 5, extract the millimeter wave radar data in the data packet, extract the Doppler speed and azimuth angle information from the millimeter wave radar data, calculate the mapping of the Doppler speed in the vehicle motion direction, obtain the vehicle motion direction speed value, calculate the vehicle motion direction speed value distribution curve, find the speed value at the peak of the distribution curve, and determine the speed value as the vehicle self-motion speed, and the specific method is as follows:

[0106] Extract the Doppler speed and azimuth angle information in the millimeter wave radar data in the data packet. Calculate the speed of each point in the vehicle motion direction using the radial speed and azimuth angle:

[0107]

[0108] Where v 车 is the vehicle motion direction speed, v 多普勒 is the Doppler speed in the millimeter wave radar data, and β is the azimuth angle in the millimeter wave radar data.

[0109] Sort v 车 , and use the Gaussian kernel function to perform kernel density estimation of v 车 to obtain the probability density function of the vehicle driving direction speed v 车 . The peak position of the probability density function corresponds to v 车 , which is the vehicle self-motion speed v ego .

[0110] Step 6, fuse the target speed and motion direction information obtained by the extended Kalman filter tracker and the vehicle self-motion speed to calculate the target absolute motion speed, and judge whether it is a moving target, and the specific steps are as follows:

[0111] Calculate the x-axis relative speed:

[0112]

[0113] Calculate the y-axis relative speed:

[0114]

[0115] Calculate the x-axis absolute speed:

[0116]

[0117] Calculate the y-axis relative speed:

[0118]

[0119] According to the speed, judge whether the target is a dynamic target:

[0120]

[0121] The application also provides an indoor environment dynamic target detection system based on a vehicle-mounted sensor system, implements the indoor environment dynamic target detection method based on the vehicle-mounted sensor system, and realizes indoor environment dynamic target detection based on the vehicle-mounted sensor system.

[0122] The application uses a self-motion estimation odometer more suitable for indoor scenes, fuses multiple sensors for target speed measurement and self-motion estimation, can well adapt to indoor scenes, reduces algorithm overhead, and ensures the real-time performance of the algorithm. At the same time, the application proposes an improved DBSCAN algorithm, improves the performance of the dynamic target detection algorithm, can well adapt to low-performance laser radars, and has great potential commercial value.

[0123] Embodiment

[0124] To verify the effectiveness of the application scheme, the following experiments are performed.

[0125] (1) Experimental preparation

[0126] The experimental vehicle hardware includes a hardware RS-LiDAR-16 line laser radar and an ARS548 laser radar. The software is deployed with Ubuntu 20.04, ros noetic, uses the improved DBSCAN algorithm for point cloud clustering, uses the extended Kalman filter and the Hungarian matching for target tracking, estimates the vehicle self-motion based on the millimeter wave radar, and fuses the sensor data to obtain the target motion state classification result.

[0127] (2) Experimental architecture

[0128] The application is based on a ROS system, subscribes original laser radar point cloud information based on ROS, pre-processes data, tracks targets and measures speed, simultaneously subscribes millimeter wave radar information, estimates motion of an unmanned vehicle, and publishes speed results through a ROS node. The accurate target motion state is obtained by fusing speed measurement data and self-motion estimation results. Figure 1 is an experimental architecture diagram.

[0129] (3) Indoor experiment

[0130] The method of the application is used in an indoor scene to detect dynamic targets, Figure 4 is a millimeter wave radar self-motion estimation result, Figure 5 is a dynamic target detection result, and red point clouds represent dynamic targets and white point clouds represent non-dynamic targets.

[0131] (4) Analysis of experimental results

[0132] Without using the application, dynamic targets in an indoor scene cannot be well perceived, which affects the accuracy and safety of unmanned driving. After using the application, accurate target tracking can be achieved in an indoor scene, millimeter wave radar data is fused for accurate self-motion estimation, and dynamic targets are identified. The application can well adapt to low-performance laser radars and complex indoor scenes, and accurate perception can be achieved under unstable light conditions.

[0133] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not exist, they should be considered as the scope of the present application.

[0134] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the present application. It should be pointed out that for ordinary skilled persons in the art, without departing from the concept of the present application, several modifications and improvements can be made, which all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method for detecting dynamic targets in an indoor environment based on a vehicle-mounted sensor system, characterized in that The method comprises the following steps: Step 1, data acquisition using a vehicle-mounted sensor system to obtain a data packet, including lidar data and millimeter wave radar data; Step 2, extracting the lidar data in the data packet, and pre-processing the lidar data, including ROI region extraction and point cloud downsampling; Step 3, pre-clustering the pre-processed lidar data using a DBSCAN algorithm, calculating the mean of the point cloud distribution variance of all clustering clusters in each direction in the pre-clustering result, and updating the Euclidean distance calculation formula according to the mean of the point cloud distribution variance in each direction, and using the updated Euclidean distance calculation formula to perform improved DBSCAN clustering on the pre-processed lidar data to obtain a final clustering cluster set of lidar targets, and the specific method is as follows: Step 3.1, preset the pre-clustering parameter neighborhood radius eps' and the minimum point number Minpts', and use the DBSCAN algorithm to pre-cluster the pre-processed laser radar data, and the pre-clustering result is represented as {M'1, M'2, …, M' n} n} m m m M' is a cluster in the pre-clustering result, which contains a plurality of laser radar points; the x-axis direction point cloud distribution variance mean of all clusters is calculated based on the coordinates of all points in each cluster M' n The calculation process is described as follows:​​​​​ Step 3.2, according to the mean value of the variance of the distribution of the point cloud in each direction The weight parameters a, b, d of the distance in the x, y, z axis direction in the Euclidean distance calculation formula are updated to obtain an improved Euclidean distance calculation formula. The specific design of the improved Euclidean distance calculation formula and the parameter calculation process are as follows: Wherein, Step 3.3, set improved DBSCAN clustering parameters neighborhood radius eps and minimum point number Minpts, in the improved DBSCAN clustering process, the distance r between the points in the preprocessed laser radar data is calculated using the improved Euclidean distance calculation formula, the two points meeting r < eps belong to the same temporary clustering cluster, the set {M1, M2, …, M n} composed of temporary clustering clusters meeting the point number greater than Minpts is the final clustering cluster set of the improved DBSCAN clustering algorithm, wherein M n ={(x1, y1, z1), (x2, y2, z2), …, (x m ,y m ,z m )}; Step 4, calculating the centroid of each clustering cluster in the final clustering cluster set of the lidar target as the position of the lidar target corresponding to the clustering cluster, constructing an extended Kalman filter tracker to track the position of the lidar target, and calculating the displacement, speed and motion direction information of the target according to the target moving track obtained by tracking; Step 5, extracting the millimeter wave radar data in the data packet, extracting the Doppler velocity and azimuth angle information from the millimeter wave radar data, calculating the mapping of the Doppler velocity in the vehicle motion direction to obtain the vehicle motion direction speed value, calculating the vehicle motion direction speed value distribution curve, and finding the speed value at the peak of the distribution curve to determine the speed value as the vehicle self-motion speed; Step 6, fusing the target speed and motion direction information obtained by the extended Kalman filter tracker and the vehicle self-motion speed to calculate the target absolute motion speed, and judging whether it is a moving target.

2. The vehicle-mounted sensor system based indoor environment dynamic target detection method of claim 1, wherein, Step 2, extracting the lidar data in the data packet, and pre-processing the lidar data, including ROI region extraction and point cloud downsampling, and the specific method is as follows: Extracting the laser radar data in the data packet, setting the ROI region extraction range threshold x according to the effective detection range of the low-beam laser radar in the indoor scene min 、 max 、 min 、 max 、 min 、 max ; Using the ROI region extraction range threshold to judge each point cloud in the lidar data, retaining the points within the ROI extraction range threshold, filtering out the points within the extraction range threshold, and obtaining the lidar data within the ROI range; Using a voxel-based downsampling method to downsample the lidar data within the ROI range to obtain the pre-processed lidar data.

3. The vehicle-mounted sensor system based indoor environment dynamic target detection method of claim 1, wherein, Step 4, calculating the centroid of each clustering cluster in the final clustering cluster set of the lidar target as the position of the lidar target corresponding to the clustering cluster, constructing an extended Kalman filter tracker to track the position of the lidar target, and calculating the speed and motion direction information of the target according to the target moving track obtained by tracking, and the specific method is as follows: Step 4.1: Traverse the final cluster set of LiDAR targets {M1, M2, ..., M} n Each cluster M in} n By obtaining the cluster M n The centroid (x, y, z) of the point set in the cluster is calculated from the mean of the x, y, and z coordinates of each point, and is used as the corresponding target location. This is then used to calculate the centroid of the point set in the cluster M. n The maximum coordinate x in the x, y, and z axes max y max z max and minimum coordinates x min y min z min The target's three-dimensional size information (l, w, h) is calculated. The specific calculation process is as follows: Step 4.2, splice the target position with the target three-dimensional size information to obtain a current time target state matrix (x, y, z, l, w, h), and perform data association between the current time target state matrix and the predicted value of the existing trajectory, and take the centroid displacement ratio and the dimension similarity as the basis for data association. If the predicted value of the existing trajectory and the centroid displacement ratio of the current time target are less than the threshold value and the dimension similarity is less than the threshold value, it is considered that the two match, otherwise it is considered not to match; the state matrix of the predicted value of the existing trajectory is (x A ,y A ,z A ,l A ,w A ,h A ), and the centroid displacement ratio is calculated in the following manner: The dimension similarity calculation method is as follows: Wherein, x_dim is the x dimension similarity, y_dim is the y dimension similarity, and z_dim is the z dimension similarity, According to the matching result, corresponding operations are performed: If the current detected pedestrian target and the existing track match successfully, the target state matrix at the current time is updated based on the extended Kalman filter update function to update the track; If the current detected pedestrian target does not match successfully, a new tracking track with (x, y, z, l, w, h) as the starting position is initialized; If the existing track is not successfully matched, the coast_cycles of the track is increased by one, and compared with the threshold, if greater than the threshold, the track is deleted; According to the matching result, the target state matrix at the current moment is updated, the tracking trajectory list is output, and the moving trajectory of the pedestrian target is output. The two-dimensional coordinates of the current detection target are (x, y), the coordinates of the target corresponding to the previous N frames of the trajectory are (x -N ,y -N ), and the displacement calculation method of the current detection target in N frames is: The speed of the current detected target is calculated in the following manner: The motion direction of the current detected target is calculated in the following manner:

4. The vehicle-mounted sensor system based indoor environment dynamic target detection method of claim 1, wherein, In step 5, millimeter wave radar data in the data packet is extracted, Doppler speed and azimuth angle information in the millimeter wave radar data are extracted, the mapping of the Doppler speed in the vehicle motion direction is calculated, the vehicle motion direction speed value is obtained, the vehicle motion direction speed value distribution curve is calculated, the speed value at the peak of the distribution curve is found, and the speed value is determined as the vehicle self-motion speed, and the specific method is as follows: Doppler speed and azimuth angle information in the millimeter wave radar data in the data packet are extracted, and the speed of each point in the vehicle motion direction is calculated using the radial speed and the azimuth angle: where v 车 is the vehicle motion direction speed, v 多普勒 is the Doppler speed in the millimeter wave radar data, and β is the azimuth angle in the millimeter wave radar data. Sort the v 车 , use the Gaussian kernel function to make the kernel density estimation of v 车 , get the probability density function of the vehicle driving direction speed v 车 , get the peak position corresponding to v 车 of the probability density function, that is, the vehicle speed v ego .

5. The vehicle-mounted sensor system based indoor environment dynamic target detection method of claim 3, wherein, In step 6, the target speed and motion direction information obtained by the extended Kalman filter tracker and the vehicle self-motion speed are fused, the target absolute motion speed is calculated, and it is judged whether it is a moving target, and the specific steps are as follows: The x-axis relative speed is calculated: The y-axis relative speed is calculated: The x-axis absolute speed is calculated: The y-axis relative speed is calculated: It is judged whether the target is a dynamic target according to the speed:

6. A vehicle-mounted sensor system-based indoor environment dynamic target detection system, characterized in that, The indoor environment dynamic target detection method based on the vehicle-mounted sensor system according to any one of claims 1-5 is implemented to realize the indoor environment dynamic target detection based on the vehicle-mounted sensor system.

Images