Target Detection Method and Device Adapted to Noise Environment

Abstract

The present invention provides a target detection method and device adaptable to a noise environment. The method includes: filtering pre-acquired point cloud data based on a preset threshold to obtain effective point cloud data; clustering the effective point cloud data to obtain effective point cloud clusters; respectively screening each cluster of effective point clouds based on a pre-acquired Gaussian distribution parameter dataset, and clustering the screened effective point clouds of each cluster based on a directed clustering algorithm to obtain target point cloud clusters, where the Gaussian distribution parameter dataset includes Gaussian distribution parameters corresponding to different distance interval features; respectively performing target detection on each cluster of target point clouds to obtain target detection results. By clustering the point clouds with abnormal features filtered out, the present invention facilitates subsequent screening of each cluster of effective point clouds in different environments and noise intensities, and then performs directed clustering, realizing the adaptability of the filtering parameters of different targets to the local environment, and reducing the false detection rate and missed detection rate.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving, and in particular, to a target detection method and device adaptable to a noise environment. Background Art

[0002] With the development of radar technology and chip technology, millimeter-wave radars are getting smaller and smaller in size, and can perform well in a multi-target complex environment, playing a better warning role for drivers, which makes people pay more and more attention to millimeter-wave radars. Due to its strong long-distance target detection ability, a millimeter-wave radar mainly measures targets in the sensor's field of view by radiating electromagnetic energy, and it still has good robustness in bad weather and strong light environments, and the cost is relatively low.

[0003] However, the complexity of the electromagnetic environment is different in different scenarios and weather conditions, so the corresponding millimeter-wave radar point cloud noise states are also different. In a strong noise environment (such as cities, tunnels, and vehicle meeting scenarios), due to the sparse real target point cloud itself and a large number of noise points, the detection performance of the millimeter-wave radar will be greatly reduced, resulting in a large number of false detections; in complex road conditions, due to the existence of a large number of false target point clouds near strong scattering surfaces such as road signs and mountains, a large number of false detections are likely to occur. In addition, the millimeter-wave radar point cloud state distributions in different environments are different, and the same set of parameters cannot be applied to various environments.

[0004] Currently, most target detections are achieved through simple preprocessing and target tracking methods. In the preprocessing process, the point cloud is filtered by setting a certain threshold. However, the setting of the threshold value is based on prior experience and cannot change according to the change of the scenario, resulting in a threshold that cannot be applied to all scenarios, and it cannot balance the miss rate and the false alarm rate, nor can it simultaneously balance the low-noise and strong-noise environments of different targets. Summary of the Invention

[0005] The present invention provides a target detection method and device adaptable to a noise environment, which solves the defect in the prior art that specific parameters cannot be applied to multiple environments, resulting in poor detection efficiency, realizes the adaptability of the filtering parameters of different targets to the local environment, and reduces the miss rate and the false alarm rate.

[0006] The present invention provides a target detection method adaptable to a noise environment, including: filtering pre-acquired point cloud data based on a preset threshold to obtain valid point cloud data; clustering the valid point cloud data to obtain valid point cloud clusters; respectively screening each cluster of valid point cloud based on a pre-acquired Gaussian distribution parameter dataset, and clustering each cluster of screened valid point cloud based on a directed clustering algorithm to obtain target point cloud clusters; wherein, the Gaussian distribution parameter dataset includes Gaussian distribution parameters corresponding to different distance interval features; respectively performing target detection on each cluster of target point cloud to obtain target detection results.

[0007] According to the target detection method adaptable to a noise environment provided by the present invention, the step of respectively screening each cluster of valid point cloud based on a pre-acquired Gaussian distribution parameter dataset includes: based on the distance features of each cluster of valid point cloud, selecting the Gaussian distribution parameters corresponding to the distance interval features to which each cluster of valid point cloud belongs; according to the Gaussian distribution parameters corresponding to each cluster of valid point cloud, screening the attribute features corresponding to each cluster of valid point cloud to obtain peak points corresponding to each cluster of valid point cloud.

[0008] According to the target detection method adaptable to a noise environment provided by the present invention, the step of clustering each cluster of screened valid point cloud based on a directed clustering algorithm includes: for each cluster of screened valid point cloud, selecting the peak point, and estimating the speed of the peak point based on least squares fitting to obtain a corresponding estimated speed; based on the estimated speed and the radial speed corresponding to the peak point, determining whether to perform least squares fitting on the screened valid point cloud of the corresponding cluster; based on performing least squares fitting on the screened valid point cloud of the corresponding cluster, obtaining a predicted speed; according to the predicted speed, determining the speed direction, and performing directed clustering according to the speed direction.

[0009] According to the target detection method adaptable to a noise environment provided by the present invention, the step of determining whether to perform least squares fitting on the screened valid point cloud of the corresponding cluster based on the estimated speed and the radial speed corresponding to the peak point includes: obtaining a difference according to the estimated speed and the radial speed corresponding to the peak point; based on the difference being greater than a preset threshold, skipping the least squares fitting of the screened valid point cloud of the corresponding cluster; otherwise, performing least squares fitting based on the screened valid point cloud.

[0010] According to the target detection method adaptable to a noise environment provided by the present invention, the step of respectively screening each cluster of valid point cloud based on a pre-acquired Gaussian distribution parameter dataset further includes: adjusting the preset threshold according to the selected Gaussian distribution parameters and the attribute features of the corresponding cluster of valid point cloud, and filtering the corresponding cluster of valid point cloud using the adjusted threshold.

[0011] A target detection method adaptable to a noise environment provided by the present invention, before filtering pre-acquired point cloud data based on a preset threshold, includes: obtaining a Gaussian distribution parameter data set, where the Gaussian distribution parameter data set includes Gaussian distribution parameters corresponding to different distance interval features; obtaining the distance feature of the point cloud data, and determining the corresponding Gaussian distribution parameter according to the distance feature of the point cloud data; and selecting a corresponding threshold as the preset threshold according to the determined Gaussian distribution parameter and the attribute feature of the point cloud data.

[0012] A target detection method adaptable to a noise environment provided by the present invention, before filtering pre-acquired point cloud data based on a preset threshold, further includes: obtaining the original point cloud data within a preset time period; and performing ego-vehicle motion compensation on each frame of the original point cloud according to the moment corresponding to each frame of the original point cloud in the original point cloud data and the relative motion of the ego-vehicle corresponding to the moment, to obtain the point cloud data.

[0013] The present invention further provides a target detection device adaptable to a noise environment, including: a data filtering module, filtering pre-acquired point cloud data based on a preset threshold to obtain effective point cloud data; a clustering module, clustering the effective point cloud data to obtain effective point cloud clusters; a data processing module, respectively screening each cluster of the effective point cloud based on a pre-acquired Gaussian distribution parameter data set, and clustering each cluster of the screened effective point cloud based on a directed clustering algorithm to obtain target point cloud clusters, where the Gaussian distribution parameter data set includes Gaussian distribution parameters corresponding to different distance interval features; and a target detection module, respectively performing target detection on each cluster of the target point cloud to obtain target detection results.

[0014] The present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the steps of the target detection method adaptable to a noise environment as described in any one of the above when executing the program.

[0015] The present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored, and the computer program implements the steps of the target detection method adaptable to a noise environment as described in any one of the above when executed by a processor.

[0016] The present invention further provides a computer program product, including a computer program, and the computer program implements the steps of the target detection method adaptable to a noise environment as described in any one of the above when executed by a processor.

[0017] The object detection method and device with noise environment adaptability provided by the present invention filter the pre-acquired point cloud data to initially screen out the point clouds with abnormal features; then cluster the effective point cloud data obtained after filtering, so as to facilitate subsequent screening of each cluster of effective point clouds based on the pre-acquired Gaussian distribution parameter dataset, thereby facilitating the adaptability of the filtering parameters of different objects to the local environment according to the different environments and various noise intensities where the objects corresponding to each point cloud are located, further reducing the interference of noise points on subsequent object detection; through directed clustering, the problem that different orientation objects require different clustering parameters is better solved, avoiding the situation of over-segmentation of objects or indistinguishable side-by-side objects, and improving the accuracy of object detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 is one of the flow diagrams of the object detection method with noise environment adaptability provided by the present invention;

[0020] Figures 2 - 5 is another flow diagram of the object detection method with noise environment adaptability provided by the present invention;

[0021] Figure 6 is the structural diagram of the object detection device with noise environment adaptability provided by the present invention;

[0022] Figure 7 is the structural diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] To make the objectives, technical solutions and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments in the present invention belong to the scope of protection of the present invention.

[0024] Figure 1 The flow diagram of an object detection method with noise environment adaptability of the present invention is shown, and the method includes:

[0025] S11, filtering the pre-acquired point cloud data based on a preset threshold to obtain effective point cloud data;

[0026] S12. Cluster the valid point cloud data to obtain valid point cloud clusters;

[0027] S13. Based on the pre-acquired Gaussian distribution parameter dataset, filter each cluster of valid point clouds respectively, and cluster the filtered valid point clouds of each cluster based on the directed clustering algorithm to obtain target point cloud clusters; wherein, the Gaussian distribution parameter dataset includes Gaussian distribution parameters corresponding to different distance interval features;

[0028] S14. Perform target detection on each cluster of target point clouds respectively to obtain target detection results.

[0029] It should be noted that S1N in this specification does not represent the sequence of the target detection method for noise environment adaptation. The following specifically combines Figures 2 - 5 to describe the target detection method for noise environment adaptation of the present invention.

[0030] Step S11. Filter the pre-acquired point cloud data based on a preset threshold to obtain valid point cloud data.

[0031] In this embodiment, due to the existence of thermal noise and various active and passive interferences during the operation of the radar, there are noises / burrs in the entire map range, and thus false target point clouds are likely to be generated. Therefore, in order to remove the radar points with abnormal features, it is necessary to filter the pre-acquired point cloud data, specifically including: comparing the attribute values of each point cloud in the point cloud data with the preset threshold, and filtering the point clouds with values lower than the preset threshold. In a possible implementation manner, the obtained valid point cloud data can refer to Figure 2 by preset thresholds corresponding to different distance intervals, so as to remove a large number of noise points.

[0032] It should be added that before filtering the pre-acquired point cloud data based on the preset threshold, it includes: presetting the threshold. Specifically including: obtaining the Gaussian distribution parameter dataset, wherein the Gaussian distribution parameter dataset includes Gaussian distribution parameters corresponding to different distance interval features; obtaining the distance feature of the point cloud data; determining the corresponding Gaussian distribution parameter according to the distance feature of the point cloud data; and selecting the corresponding threshold according to the determined Gaussian distribution parameter and the attribute feature of the point cloud data.

[0033] It should be noted that selecting the corresponding threshold according to the determined Gaussian distribution parameter and the attribute feature of the point cloud data includes: statistically analyzing the attribute feature of the global point cloud data, and determining the corresponding threshold according to the determined Gaussian distribution parameter and prior experience. For example, assume that the radar cross section (RCS) of the point cloud data follows a Gaussian distribution N~(R,σ 2 r with an expectation of R and a variance of σ 2r ) Then, R - 3*σ is correspondingly selected r as the preset threshold, that is, all the point clouds with RCS less than R - 3*σ r are filtered out. Similarly, referring to the above method, the corresponding threshold value can be selected as the preset threshold to filter out all the points with signal-to-noise ratio (SNR) lower than the threshold value.

[0034] In addition, the Gaussian distribution parameter dataset can be obtained by pre-statistically analyzing the multi-dimensional features (RCS, SNR) of the millimeter-wave radar point cloud in different distance intervals. It should be noted that the multi-dimensional features of the millimeter-wave radar point cloud are independent of each other and follow a Gaussian distribution.

[0035] In an alternative embodiment, before filtering the pre-acquired point cloud data based on the preset threshold, it further includes: acquiring the point cloud data. Specifically, it includes: acquiring the original point cloud data within a preset time period; performing self-vehicle motion compensation on each frame of the original point cloud according to the moment corresponding to each frame of the original point cloud in the original point cloud data and the relative motion of the self-vehicle corresponding to the moment, to obtain the point cloud data.

[0036] For example, the current moment is t. Assuming there is a frame of data at moment t - k, and the two timestamps at moment t and moment t - k respectively correspond to a self-vehicle position, then according to the relative value of the two poses from moment t - k to the current moment t, the data of the target in the self-vehicle coordinate system at moment t - k is compensated to the self-vehicle coordinate system at the current moment t, so as to align the point cloud frame data corresponding to different radars at different body positions to the self-vehicle positions at the current moment and the moment before the point. It should be noted that the acquired original point cloud data is from the millimeter-wave radars installed around the self-vehicle body, and the preset time period can be set according to the point cloud density, which is not further limited here.

[0037] In addition, after acquiring the original point cloud data within the preset time period, it further includes: converting the original point cloud data from the radar spherical coordinates to the vehicle body Cartesian coordinates.

[0038] Step S12, clustering the valid point cloud data to obtain valid point cloud clusters.

[0039] It should be noted that when clustering the valid point cloud data, it can be performed based on undirected clustering or segmentation to obtain multiple clusters of valid point clouds. It should be noted that for each cluster of valid point clouds, the positions of the points within it are adjacent and the attributes are similar.

[0040] Step S13, based on the pre-acquired Gaussian distribution parameter dataset, screening each cluster of valid point clouds respectively, and clustering each cluster of valid point clouds after screening based on the directed clustering algorithm to obtain the target point cloud clusters; where the Gaussian distribution parameter dataset includes the Gaussian distribution parameters corresponding to different distance interval features.

[0041] In this embodiment, based on the pre-acquired Gaussian distribution parameter dataset, each cluster of valid point clouds is screened respectively, including: based on the distance features of each cluster of valid point clouds, selecting the Gaussian distribution parameters corresponding to the distance interval features to which each cluster of valid point clouds belongs; according to the Gaussian distribution parameters corresponding to each cluster of valid point clouds, screening the attribute features corresponding to each cluster of valid point clouds to obtain the peak points corresponding to each cluster of valid point clouds.

[0042] Furthermore, screening the attribute features corresponding to each cluster of valid point clouds according to the Gaussian distribution parameters corresponding to each cluster of valid point clouds to obtain the peak points corresponding to each cluster of valid point clouds includes: obtaining the expected value according to the selected Gaussian distribution parameters; comparing the expected value with the attribute features of each cluster of valid point clouds, and if the attribute feature is greater than the expected value, the corresponding point cloud is a peak point. It should be noted that screening each cluster of valid point clouds through Gaussian distribution parameters can ensure that under the condition of different local noise floors, points with relatively high local signal-to-noise ratio are accurately selected as peak points, so as to facilitate subsequent speed estimation based on peak points and improve the estimation accuracy of the estimated speed.

[0043] For example, assume that the signal-to-noise ratio (SNR) satisfies a Gaussian distribution N~(P,σ 2 p ) with an expectation of P and a variance of σ 2 p , then the points with SNR greater than P are selected as peak points. This step can also ensure that under the condition of different local noise floors, points with relatively high local signal-to-noise ratio can be selected.

[0044] In an alternative embodiment, screening each cluster of valid point clouds based on the pre-acquired Gaussian distribution parameter dataset further includes: adjusting a preset threshold according to the selected Gaussian distribution parameters and the attribute features of the corresponding cluster of valid point clouds, and filtering the corresponding cluster of valid point clouds by using the adjusted threshold. It should be noted that after screening each cluster of valid point clouds based on the pre-acquired Gaussian distribution parameter dataset, the point clouds after secondary screening are obtained. As Figure 3 shown, the black points represent the remaining points that have not been screened out, and the gray points represent the points that have been screened out. Through the above method, a large number of miscellaneous noise points of the target on the right side of the vehicle can be filtered.

[0045] Furthermore, adjusting the preset threshold according to the selected Gaussian distribution parameters and the attribute features of the corresponding cluster of valid point clouds includes: statistically analyzing the attribute features of the global point cloud of the corresponding cluster, and determining the corresponding threshold according to the determined Gaussian distribution parameters and prior experience. It should be noted that the actual size of the threshold here is different from the above preset threshold. By dynamically adjusting the threshold according to the attribute features of different clusters of valid prior point clouds, the filtering parameters for different targets can be adapted to the local environment.

[0046] For example, the expected value of the radar cross section (RCS) is R, and the variance is σ 2 The Gaussian distribution of r is N~(R,σ 2 r ), according to the attribute characteristics of the cluster of valid point clouds, the determined Gaussian distribution parameters and prior experience, the preset threshold is adjusted from R - 3σ in step S11 r to R - σ r , so as to filter out the points with RCS lower than R - σ r . By screening the RCS, it is ensured that when the noise superposition signal exists, the preset threshold can still be adjusted according to the local statistical characteristics, so as to achieve more accurate filtering; similarly, the preset threshold of the signal-to-noise ratio (SNR) can be adjusted with reference to the above steps, so as to achieve accurate filtering of the SNR.

[0047] In an alternative embodiment, clustering is performed on each cluster of valid point clouds after screening based on the directed clustering algorithm, including: for each cluster of valid point clouds after screening, a peak point is selected, and the velocity of the peak point is estimated based on least squares fitting to obtain the corresponding estimated velocity; based on the estimated velocity and the radial velocity corresponding to the peak point, it is determined whether to perform least squares fitting on the valid point clouds of the corresponding cluster after screening; based on performing least squares fitting on the valid point clouds of the corresponding cluster after screening, a predicted velocity is obtained; according to the predicted velocity, the velocity direction is determined, and directed clustering is performed according to the velocity direction.

[0048] It should be noted that, by clustering each cluster of valid point clouds after screening based on the directed clustering algorithm, the obtained target point cloud clusters are as Figure 4 shown, where the black points are the peak points, the gray points are the millimeter wave radar point clouds, and the gray elliptical shadows are the clustering ranges of single points. It can be seen that by calculating the velocity direction through the peak points, the gray clustering range is obtained, and at the same time, the outliers in the velocity fitting are removed, realizing more accurate clustering, avoiding the situation that the first two vehicles in the front are clustered into one target due to the close distance of the point clouds, so as to improve the accuracy of target detection in the subsequent stage.

[0049] Specifically:

[0050] First, for each cluster of valid point clouds after screening, a peak point is selected, and the velocity of the peak point is estimated based on least squares fitting to obtain the corresponding estimated velocity.

[0051] In this embodiment, the velocity estimation formula is expressed as:

[0052] v x ′cosθ0 + v y ′sinθ0 = v r0

[0053] where \(v\) x ′ represents the projection of the estimated velocity of the target characterized by the peak point in the x - direction, \(v\) y ′ represents the projection of the estimated velocity of the target characterized by the peak point in the y - direction, \(\theta_0\) represents the azimuth angle of the peak point, and \(v\) r0 represents the Doppler velocity (radial velocity) of the peak point.

[0054] It should be noted that by substituting the azimuth angle of the peak point and the corresponding radial velocity into the above formula, it is convenient to estimate the velocity of the peak points with relatively high signal - to - noise ratio, improving the estimation accuracy of the estimated velocity. In addition, through clustering of the peak points, the estimated velocity direction of the point cloud cluster is regressed, and then the clustering direction is determined according to the velocity direction, which is further convenient for fine - tuning the clustering parameters to perform directed clustering, solving the problem that different - orientation targets cannot apply the same set of clustering parameters.

[0055] Secondly, based on the estimated velocity and the radial velocity corresponding to the peak point, it is judged whether to perform least - squares fitting on the effective point cloud after screening for the corresponding cluster. It should be noted that judging whether to perform least - squares fitting on the effective point cloud after screening for the corresponding cluster based on the estimated velocity and the radial velocity corresponding to the peak point includes: obtaining the difference according to the estimated velocity and the radial velocity corresponding to the peak point; skipping the least - squares fitting of the effective point cloud after screening for the corresponding cluster based on the difference being greater than the preset threshold; otherwise, performing least - squares fitting based on the screened effective point cloud.

[0056] Secondly, based on performing least - squares fitting on the effective point cloud after screening for the corresponding cluster, the predicted velocity is obtained. In this embodiment, the predicted velocity includes the projection of the true velocity of the target characterized by the corresponding point cloud cluster in the x - direction and the projection in the y - direction, and the least - squares fitting is expressed as:

[0057]

[0058] where \(v\) x represents the projection of the true velocity of the target characterized by this point cloud cluster in the x - direction, \(v\) y represents the projection of the true velocity of the target characterized by this point cloud cluster in the y - direction, \(\theta\) represents the azimuth angle of this point cloud, and \(v\) r represents the Doppler velocity (radial velocity) of this point cloud, and the superscripts \(1,2,\cdots,n\) represent the labels of the point cloud.

[0059] Finally, according to the predicted velocity, the velocity direction is determined, and directed clustering is performed according to the velocity direction. It should be noted that determining the velocity direction according to the predicted velocity includes: the vector sum of the projection of the true velocity of the target characterized by the point cloud cluster in the x - direction and the projection in the y - direction is the velocity direction.

[0060] Step S14: Perform object detection on each cluster of target point clouds respectively to obtain object detection results.

[0061] It should be noted that performing object detection on each cluster of target point clouds respectively includes: for each cluster of target point clouds, generating an object bounding box to obtain object detection results. Refer to Figure 5 , in the figure, the solid gray box represents the object bounding box, and the dashed gray box represents the detection results of the traditional method. It can be seen that due to the existence of a wall near the object on the right side of the vehicle, there are a large number of noise points near the object. Assume that there are two vehicles with different speeds in front of the vehicle and the distance is relatively close. The noise is relatively small compared to the vehicle on the right side of the vehicle, but the point clouds of the two vehicles are close in the x and y dimensions and are not easy to distinguish; at the same time, there are also many noise points in the non-object area. Detection by traditional methods is very likely to cause false detection and problems such as incorrect object position and size.

[0062] In summary, in the embodiment of the present invention, the pre-processed point cloud data is filtered to initially screen out the point clouds with abnormal features; then the effective point cloud data obtained after filtering is clustered to facilitate subsequent screening of each cluster of effective point clouds based on the pre-acquired Gaussian distribution parameter dataset, so as to facilitate the adaptation of the filtering parameters of different objects to the local environment according to the different environments and various noise intensities corresponding to each point cloud, further reducing the interference of noise points on subsequent object detection; through directed clustering, the problem that different clustering parameters are required for objects with different orientations is better solved, avoiding the situation of over-segmentation of objects or indistinguishable side-by-side objects, and improving the accuracy of object detection.

[0063] Next, the object detection device with noise environment adaptability provided by the present invention will be described. The object detection device with noise environment adaptability described below can be mutually referred to with the object detection method with noise environment adaptability described above.

[0064] Figure 6 The structural schematic diagram of an object detection device with noise environment adaptability is shown. The device includes:

[0065] A data filtering module 61 filters the pre-acquired point cloud data based on a preset threshold to obtain effective point cloud data;

[0066] A clustering module 62 clusters the effective point cloud data to obtain effective point cloud clusters;

[0067] A data processing module 63 screens each cluster of effective point clouds respectively based on the pre-acquired Gaussian distribution parameter dataset, and clusters the screened effective point clouds of each cluster based on a directed clustering algorithm to obtain target point cloud clusters; wherein, the Gaussian distribution parameter dataset includes Gaussian distribution parameters corresponding to different distance interval features;

[0068] The target detection module 64 performs target detection on each cluster of target point clouds respectively to obtain target detection results.

[0069] In this embodiment, due to the existence of thermal noise and various active and passive interferences during the operation of the radar, there are noises / burrs in the entire map range, which are likely to generate false target point clouds. Therefore, in order to remove the radar points with abnormal features, it is necessary to filter the pre-acquired point cloud data. The data filtering module 61 includes: a data filtering unit that compares the attribute values of each point cloud in the point cloud data with a preset threshold and filters out the point clouds with values lower than the preset threshold.

[0070] In an alternative embodiment, the data filtering module 61 further includes: a threshold setting unit that presets the threshold. More specifically, the threshold setting subunit includes: a data acquisition unit that acquires a Gaussian distribution parameter data set, where the Gaussian distribution parameter data set includes Gaussian distribution parameters corresponding to different distance interval features; a feature acquisition subunit that acquires the distance feature of the point cloud data; a parameter determination subunit that determines the corresponding Gaussian distribution parameter according to the distance feature of the point cloud data; and a threshold selection subunit that selects the corresponding threshold according to the determined Gaussian distribution parameter.

[0071] Among them, the threshold selection subunit includes: a first feature statistics grandson unit that statistically analyzes the attribute features of the global point cloud data; and a first threshold selection grandson unit that determines the corresponding threshold according to the determined Gaussian distribution parameter and prior experience.

[0072] In an alternative embodiment, the device further includes: a point cloud data acquisition module that acquires point cloud data. Specifically, the point cloud data acquisition module includes: a point cloud data acquisition unit that acquires the original point cloud data within a preset time period; and a vehicle motion compensation unit that performs vehicle motion compensation on each frame of the original point cloud according to the time corresponding to each frame of the original point cloud in the original point cloud data and the relative motion of the vehicle corresponding to the time, so as to obtain the point cloud data.

[0073] More specifically, for example, the current time is t. Assume that there is a frame of data at time t - k, and the two timestamps at time t and time t - k respectively correspond to a vehicle position. Then, according to the relative value of the two poses from time t - k to the current time t, the data of the target in the vehicle coordinate system at time t - k is compensated to the vehicle coordinate system at the current time t, so as to align the point cloud frame data corresponding to different times of the radars at different vehicle positions to the vehicle positions at the current time and the time before the point.

[0074] In addition, the point cloud data acquisition module further includes: a coordinate conversion unit that converts the original point cloud data from the radar spherical coordinates to the vehicle body Cartesian coordinate system.

[0075] The data processing module 63 includes: a screening unit that screens each cluster of valid point clouds based on a pre-acquired Gaussian distribution parameter dataset, where the Gaussian distribution parameter dataset includes Gaussian distribution parameters corresponding to different distance interval features; and a directed clustering unit that clusters the screened valid point clouds of each cluster based on a directed clustering algorithm to obtain target point cloud clusters.

[0076] The screening unit includes: a parameter selection subunit that selects the Gaussian distribution parameters corresponding to the distance interval features to which each cluster of valid point clouds belongs based on the distance features of each cluster of valid point clouds; and a first screening subunit that screens the attribute features corresponding to each cluster of valid point clouds according to the Gaussian distribution parameters corresponding to each cluster of valid point clouds to obtain the peak points corresponding to each cluster of valid point clouds.

[0077] Furthermore, the first screening subunit includes: an expected value acquisition sub-subunit that obtains an expected value according to the selected Gaussian distribution parameter; and a comparison sub-subunit that compares the expected value with the attribute features of each cluster of valid point clouds. If the attribute feature is greater than the expected value, the corresponding point cloud is a peak point. It should be noted that screening each cluster of valid point clouds through Gaussian distribution parameters ensures that, in the case of different local noise floors, points with relatively high local signal-to-noise ratios are accurately selected as peak points, thereby facilitating subsequent speed estimation based on peak points and improving the estimation accuracy of the estimated speed.

[0078] In an alternative embodiment, the screening unit further includes: an adjustment subunit that adjusts a preset threshold according to the selected Gaussian distribution parameter and the attribute features of the corresponding cluster of valid point clouds; and a second screening subunit that filters the corresponding cluster of valid point clouds using the adjusted threshold. Furthermore, the adjustment subunit includes: a second feature statistics sub-subunit that statistics the attribute features of the global point cloud of the corresponding cluster; and a second threshold determination sub-subunit that determines the corresponding threshold according to the determined Gaussian distribution parameter and prior experience. It should be noted that in the actual setting process, the second feature statistics sub-subunit and the first feature statistics sub-subunit can be the same unit. Similarly, the second threshold determination sub-subunit and the first threshold determination sub-subunit can also be the same unit.

[0079] In addition, the directed clustering unit includes: a speed estimation subunit that selects peak points for each cluster of screened valid point clouds and estimates the speed of the peak points based on least squares fitting to obtain the corresponding estimated speed; a judgment subunit that determines whether to perform least squares fitting on the screened valid point clouds of the corresponding cluster based on the estimated speed and the radial speed corresponding to the peak point; a speed prediction subunit that obtains a predicted speed based on performing least squares fitting on the screened valid point clouds of the corresponding cluster; and a directed clustering subunit that determines the speed direction according to the predicted speed and performs directed clustering according to the speed direction.

[0080] Specifically, the determination subunit includes: a difference obtaining grandson subunit, which obtains a difference according to the estimated speed and the radial speed corresponding to the peak point; a determination grandson subunit, which skips the least squares fitting of the effectively filtered point cloud of the corresponding cluster based on the difference being greater than a preset threshold; otherwise, performs least squares fitting based on the effectively filtered point cloud.

[0081] In summary, in the embodiment of the present invention, the data filtering module filters the pre-processed point cloud data to initially screen out the point clouds with abnormal features; then, the clustering module clusters the effectively filtered point cloud data to facilitate subsequent screening of each cluster of effectively filtered point clouds by the data processing module based on the pre-obtained Gaussian distribution parameter dataset, so as to facilitate the adaptation of the filtering parameters of different targets to the local environment according to the different environments and various noise intensities corresponding to the targets of each point cloud, further reducing the interference of noise points on subsequent target detection; the data processing module performs directed clustering to better solve the problem that different orientation targets require different clustering parameters, avoiding the situation of over-segmentation of targets or indistinguishable side-by-side targets, and improving the accuracy of target detection.

[0082] Figure 7 An example of the physical structure diagram of an electronic device is shown as Figure 7 As shown, the electronic device may include: a processor 71, a communication interface 72, a memory 73, and a communication bus 74. Among them, the processor 71, the communication interface 72, and the memory 73 complete mutual communication through the communication bus 74. The processor 71 can call the logical instructions in the memory 73 to execute the target detection method for noise environment adaptation, and the method includes: filtering the pre-obtained point cloud data based on a preset threshold to obtain effectively filtered point cloud data; clustering the effectively filtered point cloud data to obtain effectively filtered point cloud clusters; respectively screening each cluster of effectively filtered point clouds based on the pre-obtained Gaussian distribution parameter dataset, and clustering each cluster of effectively filtered point clouds after screening based on a directed clustering algorithm to obtain target point cloud clusters; where the Gaussian distribution parameter dataset includes Gaussian distribution parameters corresponding to different distance interval features; respectively performing target detection on each cluster of target point clouds to obtain target detection results.

[0083] In addition, when the logical instructions in the above-mentioned memory 73 can be implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0084] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the noise environment adaptive target detection method provided by the above-mentioned various methods. The method includes: filtering the pre-acquired point cloud data based on a preset threshold to obtain effective point cloud data; clustering the effective point cloud data to obtain effective point cloud clusters; based on the pre-acquired Gaussian distribution parameter dataset, screening each cluster of effective point clouds respectively, and clustering the screened effective point clouds of each cluster based on a directed clustering algorithm to obtain target point cloud clusters; wherein, the Gaussian distribution parameter dataset includes Gaussian distribution parameters corresponding to different distance interval features; performing target detection on each cluster of target point clouds respectively to obtain target detection results.

[0085] In yet another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the noise environment adaptive target detection method provided by the above-mentioned various methods. The method includes: filtering the pre-acquired point cloud data based on a preset threshold to obtain effective point cloud data; clustering the effective point cloud data to obtain effective point cloud clusters; based on the pre-acquired Gaussian distribution parameter dataset, screening each cluster of effective point clouds respectively, and clustering the screened effective point clouds of each cluster based on a directed clustering algorithm to obtain target point cloud clusters; wherein, the Gaussian distribution parameter dataset includes Gaussian distribution parameters corresponding to different distance interval features; performing target detection on each cluster of target point clouds respectively to obtain target detection results.

[0086] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

[0087] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A target detection method adaptable to a noise environment, characterized in that, Including: Filtering the pre-acquired point cloud data based on a preset threshold to obtain effective point cloud data; Clustering the effective point cloud data to obtain effective point cloud clusters; Based on a pre-acquired Gaussian distribution parameter dataset, screening each cluster of effective point clouds respectively, and clustering the screened effective point clouds of each cluster based on a directed clustering algorithm to obtain target point cloud clusters; wherein, the Gaussian distribution parameter dataset includes Gaussian distribution parameters corresponding to different distance interval features; Performing target detection on each cluster of target point clouds respectively to obtain target detection results; The clustering of the screened effective point clouds of each cluster based on the directed clustering algorithm includes: For each cluster of screened effective point clouds, selecting peak points, and estimating the speed of the peak points based on least squares fitting to obtain corresponding estimated speeds; Based on the estimated speed and the radial speed corresponding to the peak points, determining whether to perform least squares fitting on the screened effective point clouds of the corresponding cluster; Based on performing least squares fitting on the screened effective point clouds of the corresponding cluster, obtaining a predicted speed; According to the predicted speed, determining the speed direction, and performing directed clustering according to the speed direction.

2. The target detection method adaptable to a noise environment according to claim 1, characterized in that, The screening of each cluster of effective point clouds respectively based on the pre-acquired Gaussian distribution parameter dataset includes: Based on the distance features of each cluster of effective point clouds, selecting the Gaussian distribution parameters corresponding to the distance interval features to which each cluster of effective point clouds belongs; According to the Gaussian distribution parameters corresponding to each cluster of effective point clouds, screening the attribute features corresponding to each cluster of effective point clouds to obtain peak points corresponding to each cluster of effective point clouds.

3. The target detection method adaptable to a noise environment according to claim 1, characterized in that, The determining whether to perform least squares fitting on the screened effective point clouds of the corresponding cluster based on the estimated speed and the radial speed corresponding to the peak points includes: Obtaining a difference according to the estimated speed and the radial speed corresponding to the peak points; Based on the difference being greater than a preset threshold, skipping the least squares fitting of the screened effective point clouds of the corresponding cluster; otherwise, performing least squares fitting based on the screened effective point clouds.

4. The target detection method adaptable to a noise environment according to claim 2, characterized in that, The screening of each cluster of effective point clouds respectively based on the pre-acquired Gaussian distribution parameter dataset further includes: Adjusting the preset threshold according to the selected Gaussian distribution parameters and the attribute features of the corresponding cluster of effective point clouds, and filtering the corresponding cluster of effective point clouds using the adjusted threshold.

5. The target detection method adaptable to a noise environment according to claim 1, characterized in that, Before filtering the pre-acquired point cloud data based on a preset threshold, it includes: Obtaining a Gaussian distribution parameter dataset, wherein the Gaussian distribution parameter dataset includes Gaussian distribution parameters corresponding to different distance interval features; Obtaining the distance features of the point cloud data, and determining the corresponding Gaussian distribution parameters according to the distance features of the point cloud data; Selecting a corresponding threshold as the preset threshold according to the determined Gaussian distribution parameters and the attribute features of the point cloud data.

6. The target detection method adaptable to a noise environment according to claim 1, characterized in that, Before filtering the pre-acquired point cloud data based on a preset threshold, it further includes: Obtaining the original point cloud data within a preset time period; Perform ego-vehicle motion compensation on each frame of the original point cloud according to the time corresponding to each frame of the original point cloud in the original point cloud data and the relative motion of the ego-vehicle corresponding to the time, to obtain point cloud data.

7. A target detection device adaptable to a noise environment, characterized in that, Including: A data filtering module, which filters the pre-acquired point cloud data based on a preset threshold to obtain valid point cloud data; A clustering module, which clusters the valid point cloud data to obtain valid point cloud clusters; A data processing module, which respectively screens each cluster of valid point clouds based on a pre-acquired Gaussian distribution parameter dataset, and clusters each cluster of valid point clouds after screening based on a directed clustering algorithm to obtain target point cloud clusters; wherein, the Gaussian distribution parameter dataset includes Gaussian distribution parameters corresponding to different distance interval features; A target detection module, which respectively performs target detection on each cluster of target point clouds to obtain target detection results; The data processing module includes: a directed clustering unit, which is used to cluster each cluster of valid point clouds after screening based on a directed clustering algorithm to obtain target point cloud clusters; The directed clustering unit includes: A speed estimation sub-unit, which, for each cluster of valid point clouds after screening, selects peak points and estimates the speed of the peak points based on least squares fitting to obtain corresponding estimated speeds; A judgment sub-unit, which judges whether to perform least squares fitting on the valid point clouds of the corresponding cluster after screening based on the estimated speed and the radial speed corresponding to the peak points; A speed prediction sub-unit, which obtains a predicted speed based on performing least squares fitting on the valid point clouds of the corresponding cluster after screening; A directed clustering sub-unit, which determines a speed direction according to the predicted speed and performs directed clustering according to the speed direction.

8. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the target detection method for noise environment adaptation according to any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the target detection method for noise environment adaptation according to any one of claims 1 to 6.

Images