Target detection method and device

Abstract

The invention relates to a target detection method and device, and the method comprises the steps: obtaining N continuous frames of point clouds collected by a laser radar, and N is a positive integer greater than or equal to 2; determining one frame of point cloud in the N frames of point cloud as a target frame of point cloud; aligning (N-1) frames of point clouds except the target frame of point clouds in the N frames of point clouds to the target frame of point clouds, and obtaining the positions of points in the (N-1) frames of aligned point clouds and any frame of aligned point clouds as estimated positions of the points in the point clouds before alignment at the moment corresponding to the target frame of point clouds; fusing the (N-1)-frame aligned point cloud with the target frame point cloud to obtain a fused frame point cloud; and performing target detection on the fused frame point cloud to obtain a target detection result corresponding to the N frame point cloud. According to the method, densification of the point cloud can be realized, target detection is carried out based on the densified fusion frame point cloud, and the detection rate of target detection is improved.

Description

Technical Field

[0001] The present application belongs to the technical field of target detection based on point cloud data, and in particular, to a target detection method and device. Background Art

[0002] LiDAR is a radar system that uses laser beams to detect the position, speed and other characteristic quantities of a target. Its working principle is to transmit a detection signal to the target, and then compare the received signal reflected from the target with the transmitted signal to obtain information such as the distance, position, speed, and shape of the target. LiDAR point cloud is a series of discrete point data obtained by LiDAR during the measurement process. LiDAR point cloud can be used for target detection and tracking.

[0003] The technology of using LiDAR point cloud for target detection is widely used in fields such as autonomous driving and mobile robots. In the process of target detection, the detection conditions such as the size of the target to be detected and the distance between the position of the target to be detected and the LiDAR are diverse and often change. The point cloud generated by LiDAR at a long distance or for a small target to be detected is very sparse. Sparse point clouds have poor performance for target features. Therefore, when using sparse point clouds for target detection, it is often difficult to extract target features from sparse point clouds, resulting in low target detection rate and high false detection rate.

[0004] Therefore, how to improve the detection rate of target detection for sparse point clouds and thus improve the detection effect has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The embodiments of the present application provide a target detection method and device, which can solve the technical problem in the prior art that the detection rate of target detection for sparse point clouds is low, thereby resulting in cross-target detection effects.

[0006] In a first aspect, an embodiment of the present application provides a target detection method, which includes: obtaining N consecutive frames of point clouds collected by a laser radar, where N is a positive integer greater than or equal to 2; determining one frame of point cloud in the N frames of point cloud as a target frame point cloud; aligning N-1 frames of point cloud other than the target frame point cloud in the N frames of point cloud to the target frame point cloud, respectively, to obtain N-1 frames of aligned point cloud, the position of a point in any frame of the aligned point cloud being the estimated position of the point in the point cloud before alignment at the corresponding moment of the target frame point cloud; fusing the N-1 frames of aligned point cloud with the target frame point cloud to obtain a fused frame point cloud; performing target detection on the fused frame point cloud to obtain a target detection result corresponding to the N frames of point cloud.

[0007] In the above method, one frame of point cloud in the multi-frame point cloud is determined as the target frame point cloud, and the other frame point clouds in the multi-frame point cloud are aligned with the target frame point cloud. By aligning the point clouds, the estimated positions of the points in the other frame point clouds at the time corresponding to the target frame point cloud are obtained, and then the aligned point clouds after the other frame point clouds are aligned to the target frame point cloud are obtained, and then the aligned point clouds of the other frame point clouds are fused with the target frame point cloud to obtain the fused frame point cloud, thereby achieving the densification of the point cloud; and then the target detection is performed based on the densified fused frame point cloud to improve the detection rate of the target detection. In addition, when there are points corresponding to different positions of the same target to be detected in different frame point clouds, the information of the target to be detected between different frame point clouds can be supplemented by obtaining the fused frame point cloud, thereby improving the detection rate and detection accuracy of the target to be detected.

[0008] In one embodiment, N-1 frames of point clouds other than the target frame point cloud in the N frames of point cloud are aligned to the target frame point cloud to obtain the N-1 frames of aligned point clouds, including:

[0009] According to the first frame point cloud and the target frame point cloud, determining position conversion information corresponding to the first frame point cloud, the position conversion information including position change parameters of points in the first frame point cloud converted to estimated positions, the first frame point cloud being any one frame point cloud in the N-1 frame point clouds;

[0010] The points in the first frame point cloud are transformed using the position transformation information to obtain an aligned point cloud corresponding to the first frame point cloud.

[0011] In one embodiment, the position conversion information includes a motion vector, and determining the position conversion information corresponding to the first frame point cloud according to the first frame point cloud and the target frame point cloud includes:

[0012] The first frame point cloud and the target frame point cloud are input into a preset scene flow estimation algorithm to obtain a motion vector corresponding to the first frame point cloud.

[0013] In one embodiment, the motion vector corresponding to the first frame point cloud includes a coordinate transformation vector corresponding to each point in the first frame point cloud; using the position transformation information, transforming the points in the first frame point cloud to obtain an aligned point cloud corresponding to the first frame point cloud, including:

[0014] Add each point in the first frame point cloud to the corresponding coordinate transformation vector to obtain the aligned point cloud corresponding to the first frame point cloud.

[0015] In one embodiment, the position conversion information includes a transformation matrix, and determining the position conversion information corresponding to the first frame point cloud according to the first frame point cloud and the target frame point cloud includes:

[0016] The first frame point cloud and the target frame point cloud are input into a preset point cloud registration algorithm to obtain a transformation matrix corresponding to the first frame point cloud.

[0017] In one embodiment, the data collected by the laser radar includes M consecutive frames of point clouds, and the N frame point cloud is any consecutive N frame point cloud in the M frame point cloud, where M is a positive integer greater than N.

[0018] In one embodiment, the target frame point cloud is the Kth frame point cloud in time sequence among N frame point clouds, where 1≤K≤N and K is a positive integer.

[0019] In one embodiment, the target frame point cloud is a frame point cloud having the largest number of points among N frames point clouds.

[0020] In one embodiment, the laser radar is a roadside laser radar, and the N-frame point clouds are point clouds obtained by the laser radar scanning the same scanning area at different times; the scanning area includes an area of ​​interest and an area of ​​non-interest, the area of ​​interest is the area where the target to be detected appears, the target frame point cloud is a frame point cloud that includes the most target points in the N-frame point clouds, and the target point is a point in the point cloud located in the area of ​​interest.

[0021] In a second aspect, an embodiment of the present application provides a target detection device, which includes a unit for executing each step of the method as described in any one of the first aspects above.

[0022] In a third aspect, an embodiment of the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the methods described in the first aspect when executing the computer program.

[0023] In a fourth aspect, an embodiment of the present application provides a target detection system, which includes a laser radar and a computer device as described in any one of the second aspects above.

[0024] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in any one of the above first aspects is implemented.

[0025] In a sixth aspect, an embodiment of the present application provides a chip, comprising: a processor, configured to call and run a computer program from a memory, so that a computer device equipped with the chip executes a method as described in any one of the above first aspects.

[0026] It can be understood that the beneficial effects of the second to sixth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0028] Figure 1 is a schematic diagram of the structure of a target system provided by an embodiment of the present application;

[0029] Figure 2 is a flowchart of a target detection method provided by an embodiment of the present application;

[0030] Figure 3 It is a schematic diagram of the principle of using a scene flow estimation algorithm to perform point cloud alignment in a target detection method provided in an embodiment of the present application;

[0031] Figure 4 is a schematic diagram of a point cloud alignment and fusion process in a target detection method provided in an embodiment of the present application;

[0032] Figure 5 is a structural block diagram of a target detection device provided by an embodiment of the present application;

[0033] Figure 6 It is an internal structure diagram of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0034] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0035] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0036] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0037] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0038] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0039] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0040] The embodiment of the present application provides a target detection method, which obtains continuous multi-frame point clouds collected by a laser radar, and determines one frame point cloud in the multi-frame point cloud as a target frame point cloud, and then aligns other frame point clouds in the multi-frame point cloud with the target frame point cloud (that is, obtains the estimated position of the points in other frame point clouds at the time corresponding to the target frame point cloud, and obtains the aligned point cloud after the other frame point clouds are aligned to the target frame point cloud), and then fuses the aligned point cloud with the target frame point cloud to obtain a fused frame point cloud, and performs target detection on the basis of the fused frame point cloud. The method aligns the point clouds, and then fuses the aligned point clouds of other frame point clouds with the target frame point cloud to obtain a fused frame point cloud, thereby achieving densification of the point cloud; performing target detection based on the densified fused frame point cloud can improve the detection rate of target detection and improve the accuracy of target detection. In addition, when there are points corresponding to different positions of the same target to be detected in different frame point clouds, by obtaining the fused frame point cloud, the information of the target to be detected between different frame point clouds can be supplemented, thereby improving the detection rate and detection accuracy of the target to be detected.

[0041] The target detection method provided in the present application is exemplarily described below in conjunction with specific embodiments.

[0042] See also Figure 1, is a schematic diagram of the structure of a target system provided in an embodiment of the present application, such as Figure 1 As shown, the target detection system includes a computing device 110 and a laser radar 120, wherein the computing device 110 and the laser radar 120 are communicatively connected via a communication interface, wherein the communication interface may be a wired communication interface or a wireless communication interface.

[0043] The laser radar 120 is used to collect point clouds in the target area to obtain multi-frame point clouds, and send the obtained point clouds to the computer device 110. The computer device 110 aligns some frame point clouds in the multi-frame point clouds to the target frame point cloud, and fuses the aligned point cloud with the target frame point cloud to obtain a fused frame point cloud, and finally performs target detection on the fused frame point cloud to obtain a target detection result.

[0044] It is understandable that the laser radar 120 in the embodiment of the present application can be either a roadside laser radar or a vehicle-mounted laser radar; and the computer device 110 can be any device that has target detection task requirements. For example, the computer device 110 can be a road-side computer device, a vehicle-side computer device, or a combination system of a road-side computer device and a vehicle-side computer device. This application does not limit this.

[0045] Exemplarily, the laser radar 120 may be a roadside laser radar, for example, the laser radar 120 may be a laser radar in a roadside fusion perception system. In this case, the computing device 110 may be a road-side computer device, for example, a computing unit in a roadside fusion perception system (also known as a smart base station or a roadside base station), for example, an industrial computer in a roadside fusion perception system.

[0046] It should be understood that the target detection system in the embodiment of the present application can be applied to any scenario in which target detection is performed based on a lidar point cloud. For example, the lidar 120 and the computer device 110 can also be an integrated target detection system, and the present application does not impose any restrictions on this.

[0047] For example, taking the laser radar 120 as a roadside laser radar, the laser radar user detects pedestrians on the road. If the pedestrian is far away from the laser radar 120, the point cloud formed by the laser radar at the pedestrian is relatively sparse, and the target detection accuracy based on the sparse point cloud is low; in addition, pedestrians are moving on the road. In different frame point clouds, a certain frame point cloud may be a point cloud at the shoulder position of the pedestrian, while the next frame point cloud may be a point cloud at the pedestrian's legs and other positions. By obtaining a fused frame point cloud, the detection target information between different frame point clouds can be supplemented, thereby improving the detection rate and detection accuracy of the detection target.

[0048] Based on Figure 1 The target detection system shown in the present application provides, in one embodiment, Figure 2 The target detection method shown in the figure is described by taking the method applied to the above-mentioned computer device 110 as an example. It can be understood that the following description is only an example and does not constitute a limitation on the protection scope of the present application. Figure 2 As shown, the method may include S201 to S205. Each step is described below.

[0049] Step S201: Obtain N consecutive frames of point clouds collected by the laser radar, where N is a positive integer greater than or equal to 2.

[0050] In the embodiment, N continuous frames of point cloud refer to N continuous frames of point cloud in the acquisition time of the laser radar. Assuming that the laser radar acquires one frame of point cloud per second, then the N frames of point cloud are the point clouds acquired within N seconds.

[0051] In the embodiment of the present application, N frames of point cloud are processed as a group of data, where the size of N depends on factors such as the complexity of the target detection task. Technical personnel in this field can adjust it as needed, and the embodiment of the present application does not limit this.

[0052] It is understandable that when the laser radar collects point clouds in the detection area, the collection usually lasts for a long time. Therefore, the data collected by the laser radar may include many frames of point clouds. The N in the N frame point cloud is a set of frames for processing data as needed.

[0053] In some embodiments, the data collected by the laser radar includes M continuous frames of point clouds, and the N frame point cloud is any continuous N frame point cloud in the M frame point cloud, where M is a positive integer greater than N. In this embodiment, the N frame point cloud is any continuous N frame point cloud in the M frame point cloud, that is, the M frame point cloud can be grouped and processed, and each group of point clouds includes continuous N frame point clouds. The same frame point clouds may or may not exist between different groups of point clouds. In addition, in an embodiment, the N frame point cloud is a partial point cloud of the laser radar point cloud. By combining any N frame point clouds (i.e., dividing the point cloud collected by the laser radar into multiple groups), continuous target detection can be achieved.

[0054] For example, assuming that M is 1000 and N is 10, the laser radar collects 1000 consecutive frames of point clouds, and each set of point clouds includes 10 consecutive frames of point clouds.

[0055] (1) If there is no identical frame point cloud in each set of point clouds, then in the 1000 frame point clouds:

[0056] The 1st frame point cloud to the 10th frame point cloud, a total of 10 frames of point cloud are the first group of point clouds;

[0057] The 11th frame point cloud to the 20th frame point cloud, a total of 10 frames of point cloud are the second group of point clouds;

[0058] The 21st frame point cloud to the 30th frame point cloud, a total of 10 frames of point cloud are the third group of point clouds;

[0059] …

[0060] The 991st frame point cloud to the 1000th frame point cloud, a total of 10 frames of point clouds, are the 100th group of point clouds.

[0061] In this case, 1000 frames of point clouds can be divided into 100 groups of point clouds in total, each group of point clouds includes 10 consecutive frames of point clouds, and the operations from step S202 to step S205 are performed on each group of point clouds.

[0062] (2) If there are points in the same frame in each point cloud group, then in 1000 point clouds:

[0063] The 1st frame point cloud to the 10th frame point cloud, a total of 10 frames of point cloud are the first group of point clouds;

[0064] The 2nd frame point cloud to the 11th frame point cloud, a total of 10 frames of point cloud, are the second group of point clouds;

[0065] The point cloud from the 3rd frame to the 12th frame, a total of 10 frames of point cloud, is the third group of point cloud;

[0066] …

[0067] The 990th frame point cloud to the 999th frame point cloud, a total of 10 frames of point cloud, is the 990th group of point clouds.

[0068] The 991st frame point cloud to the 1000th frame point cloud, a total of 10 frames of point clouds, are the 991th group of point clouds.

[0069] In this case, 1000 frames of point clouds can be divided into 991 groups of point clouds in total, each group of point clouds includes 10 consecutive frames of point clouds, and the operations from step S202 to step S205 are performed on each group of point clouds.

[0070] It is understandable that the above method for obtaining N frames of point clouds from M frames of point clouds is only an example and is not intended to limit the scope of protection of this application. Those skilled in the art may also group M frames of point clouds in other ways, which are not enumerated here.

[0071] Step S202: determine one frame of point cloud among the N frames of point cloud as the target frame of point cloud.

[0072] In an embodiment, after obtaining N frames of point clouds, one frame of point cloud is determined from the N frames of point clouds as a target frame of point cloud, wherein the target frame of point cloud serves as a reference frame of point cloud in subsequent point cloud alignment, that is, in subsequent point cloud alignment, other frame of point clouds in the N frames of point cloud need to be aligned to the target frame of point cloud.

[0073] In some embodiments, any one of the N frame point clouds can be determined as the target frame point cloud. In this method, the target frame point cloud can be randomly selected from the N frame point clouds, and the operation is relatively simple.

[0074] In some embodiments, the Kth frame point cloud in the time sequence of N frame point clouds, where 1≤K≤N and K is a positive integer, can be determined as the target frame point cloud. This selection method makes it easy to operate the N frame point clouds of different groups because the selection of the target frame point clouds has a specific regularity; in addition, this operation can ensure that the target frame point clouds of different groups of N frame point clouds are different, so that the detection results of different groups of N frame point clouds represent the positions of the detection targets at different times (i.e., the time corresponding to the target point cloud), ensuring that the target detection results are widely distributed in time and the practicality of the target detection results.

[0075] In some optional embodiments, the target frame point cloud is the frame point cloud with the largest number of points among the N frame point clouds. In this implementation, the frame point cloud with the largest number of points among the N frame point clouds (i.e., the frame point cloud with the largest point cloud density) is determined as the target frame point cloud. Since only other point clouds with fewer points need to be aligned to the target frame point cloud, the amount of calculation can be reduced; and the frame point cloud with the largest number of points is used as the reference for point cloud alignment, and the probability that other frame point clouds have corresponding points with the target frame point cloud is greater, so the accuracy of point cloud alignment can be improved.

[0076] In some optional embodiments, the laser radar is a roadside laser radar, and the N-frame point clouds are point clouds obtained by the laser radar scanning the same scanning area at different times; the scanning area includes an area of ​​interest and an area of ​​non-interest, the area of ​​interest is the area where the target to be detected appears, the target frame point cloud is a frame point cloud in the N-frame point clouds that includes the most target points, and the target point is a point in the point cloud located in the area of ​​interest.

[0077] For example, assuming that the scanning area is an intersection and the target to be detected is a traffic participant on the road (such as a pedestrian or a vehicle), then the road area is the area of ​​interest and the roadside green belt is the area of ​​non-interest; the area of ​​interest can be first determined for each frame of point cloud, and the frame of point cloud with the most points in the area of ​​interest can be used as the target frame point cloud. The definition of the area of ​​interest ensures that the selected target frame point cloud has the most useful points, thereby further improving the detection accuracy.

[0078] Step S203: Align N-1 frames of point clouds other than the target frame point cloud in the N frame point cloud to the target frame point cloud, respectively, to obtain N-1 frames of aligned point clouds. The position of a point in any frame of the aligned point cloud is the estimated position of the point in the point cloud before alignment at the corresponding moment of the target frame point cloud.

[0079] In the embodiment, each frame of point cloud corresponds to a moment, different frames of point cloud correspond to different moments, and the moment corresponding to each frame of point cloud may be the acquisition moment of the frame of point cloud.

[0080] Assume that any frame of point cloud in N-1 frames is point cloud A, which includes 50 points and the acquisition time of the point cloud is t A ; The target frame point cloud is point cloud X, and the acquisition time of point cloud X is t X When point cloud A needs to be aligned to point cloud X, specifically, for each point in point cloud A, from t A to X The displacement of each point in the point cloud A is estimated within the time period of t X The estimated positions of the 50 points in point cloud A are the positions of the points in the aligned point cloud.

[0081] In some embodiments, the alignment process in step S203 specifically includes the following steps:

[0082] (1) According to the first frame point cloud and the target frame point cloud, position conversion information corresponding to the first frame point cloud is determined, wherein the position conversion information includes position change parameters of points in the first frame point cloud converted to estimated positions, and the first frame point cloud is any one frame point cloud in the N-1 frame point clouds.

[0083] In an embodiment, the position conversion information corresponding to the first frame point cloud includes position change parameters for converting points in the first frame point cloud to estimated positions. The position change parameters are used to represent the position change data of points in the first frame point cloud. The definition of the estimated position is as described above and will not be repeated here.

[0084] (2) Using the position transformation information, the points in the first frame point cloud are transformed to obtain an aligned point cloud corresponding to the first frame point cloud.

[0085] It can be understood that the alignment of the first frame point cloud and the target frame point cloud is specifically a process of dynamic target behavior estimation, that is, estimating the position of the target in the first frame point cloud at the time when the target frame point cloud is located.

[0086] It should be understood that different methods are used to achieve the alignment of the first frame point cloud and the target frame point cloud, and the position conversion information obtained will be different, and the conversion process of the first frame point cloud will also be different. The following example is used to illustrate.

[0087] In some embodiments, a scene flow estimation method is used to achieve alignment of the first frame point cloud and the target frame point cloud. In this case, the position conversion information includes a motion vector, and obtaining the position conversion information specifically includes: obtaining a motion vector corresponding to the first frame point cloud by inputting the first frame point cloud and the target frame point cloud into a preset scene flow estimation algorithm.

[0088] Exemplarily, the motion vector corresponding to the first frame point cloud includes a coordinate transformation vector corresponding to each point in the first frame point cloud, and the points in the first frame point cloud are transformed using the transformation information, specifically including: adding each point in the first frame point cloud to the corresponding coordinate transformation vector to obtain an aligned point cloud corresponding to the first frame point cloud.

[0089] Figure 3 The schematic diagram of the principle of using the scene flow estimation algorithm to perform point cloud alignment in the target detection method provided in one embodiment of the present application. Figure 3 As shown, the input of the scene flow estimate algorithm is two frames of adjacent point clouds P 1 ={p i |i=1,...,n 1}, P t ={p ti |i=1,...,n 2}, where p i and p ti is the three-dimensional coordinate (i.e. x, y, z coordinate), n 1 and n 2 They are the first frame point cloud P 1 The number of points in the target frame and the point cloud P t The number of points in the first frame can be obtained by the scene flow estimation algorithm. 1 To the target frame point cloud P t The output of the scene flow estimation algorithm is D = {d i |i=1,...,n 1}, where d i is the i-th point p in the first frame point cloud predicted by the scene flow estimation algorithm i Coordinate transformation vector to the target frame (i.e. displacement in x, y, z directions).

[0090] For example, the first frame point cloud P 1 Add it to the scene flow output D to get the point cloud P of the first frame 1 The corresponding aligned point cloud (that is, equivalent to aligning the first frame point cloud to the target frame point cloud). For example, for a point in the first frame point cloud, assume that its coordinates in the first frame point cloud are p m , the coordinate transformation vector of this point obtained by the scene flow estimation algorithm is d m , then the predicted coordinates of the point in the target frame point cloud are p m '=p m +d m .

[0091] like Figure 3As shown, in the output D of the scene flow estimation algorithm, the number of coordinate transformation vectors is equal to the number of points in the first frame point cloud. 1 , so the number of points in the point cloud after alignment is also n 1 .

[0092] In some other embodiments, a point cloud registration method may be used to align the first frame point cloud with the target frame point cloud. In this case, the position conversion information includes a change matrix, and the process of obtaining the change matrix is ​​as follows: inputting the first frame point cloud and the target frame point cloud into a preset point cloud registration algorithm to obtain a transformation matrix corresponding to the first frame point cloud.

[0093] Exemplarily, using the transformation information to transform the points in the first frame point cloud specifically includes: using a change matrix to transform the coordinates of each point in the first frame point cloud to obtain an aligned point cloud corresponding to the first frame point cloud.

[0094] Exemplarily, the point cloud registration method may be an iterative closest point algorithm (ICP) or a feature-based image registration algorithm (Feature-Based Registration):

[0095] (1) Iterative Closest Point, ICP: This algorithm is a classic point cloud registration algorithm. It iteratively searches for the best rigid transformation to minimize the distance between two point clouds. The basic principle of the ICP algorithm is to match the corresponding point pairs in the two point clouds and continuously update the rigid transformation matrix until the optimal registration result is achieved.

[0096] (2) Feature-Based Registration: This algorithm relies on the features of point clouds for registration. It usually first uses a feature extraction algorithm to extract key features of point clouds, such as surface descriptors, corners or edges, and then matches these features between two frames of point clouds to finally derive a rigid transformation. Common feature extraction algorithms include SIFT (Scale Invariant Feature Transform), SURF (Speeded Up Robust Feature), ORB (Oriented FAST and Rotated BRIEF), etc.

[0097] Step S204: Fuse the aligned point cloud of the N-1 frame with the point cloud of the target frame to obtain a fused frame point cloud.

[0098] In an embodiment, after obtaining the N-1 frame aligned point cloud, the N-1 frame point cloud is fused with the target frame point cloud. Specifically, the points in the N-1 frame aligned point cloud are superimposed on the target frame point cloud to obtain a fused frame point cloud.

[0099] Figure 4 Schematic diagram of the point cloud alignment and fusion process in the target detection method provided in one embodiment of the present application. Figure 4 As shown, in this embodiment, the scene flow estimation algorithm is used to align the point clouds, and then the aligned point clouds are superimposed on the target frame point clouds.

[0100] like Figure 4 As shown, it is assumed that n frames of point clouds are aligned and superimposed, where the n frames of point clouds are: P 1 , P 2 , P 3 ...P n-1 , P n , the point cloud P n is the target frame point cloud. 1 , P 2 , P 3 ...P n-1 , P n The number of midpoints is: n 1 、n 2 、n 3 ……n 1n-1 、n n , and each point is a three-dimensional coordinate. The specific process is as follows Figure 4 As shown:

[0101] (1) P 1 and P n Input to the scene flow estimation algorithm, the scene flow estimation algorithm outputs the motion vector D corresponding to the point cloud of the first frame 1 , then P 1 With D 1 Superposition to obtain the aligned point cloud P 1 ', where D 1 Include 1 Coordinate transformation vector, aligned point cloud P 1 'Include n 1 points.

[0102] (3) P 2 and P n Input to the scene flow estimation algorithm, the scene flow estimation algorithm outputs the motion vector D corresponding to the point cloud of the first frame 2 , then P 2 With D 2 Superposition to obtain the aligned point cloud P 2 ', where D 2 Include2 Coordinate transformation vector, aligned point cloud P 2 'Include n 2 points.

[0103] (3) P 3 and P n Input to the scene flow estimation algorithm, the scene flow estimation algorithm outputs the motion vector D corresponding to the point cloud of the first frame 3 , then P 3 With D 3 Superposition to obtain the aligned point cloud P 3 ', where D 2 Include 3 Coordinate transformation vector, aligned point cloud P 3 'Include n 3 points.

[0104] (4) And so on, until P n-1 and P n Input to the scene flow estimation algorithm, the scene flow estimation algorithm outputs the motion vector D corresponding to the point cloud of the first frame n-1 , then P n-1 With D n-1 Superposition to obtain the aligned point cloud P n-1 ', where D n-1 Include n-1 Coordinate transformation vector, aligned point cloud P n-1 'Include n n-1 points.

[0105] (5) Finally, the aligned point cloud P 1 '、P 2 '、P 3 '……P n-1 ', and P n Superposition, to obtain the fused frame point cloud, the number of points in the fused frame point cloud is n 1 +n 2 +n 3 +……+n n .

[0106] It can be understood that whether the point density in the fused frame point cloud in the embodiment of the present application can be effectively improved is closely related to the cloud alignment method and the state of the target in the point cloud. The following is an explanation using the roadside lidar as an example.

[0107] Suppose: in the n-frame point cloud, for the dynamic target G1, since the target G1 is a dynamic object relative to the laser radar, the points of the target G1 in the first frame point cloud and the target frame point cloud should be different; the aligned point cloud of the first frame point cloud obtained by the scene flow estimation algorithm, or the aligned point cloud of the first frame point cloud obtained by the point cloud registration method, the coordinates of the points in the aligned point cloud will be different from the coordinates of the points in the target frame point cloud. Therefore, in the fused frame point cloud after the aligned point cloud is fused with the target frame point cloud, the density of the point cloud will be greater than the point density in the target frame point cloud, thereby realizing the densification of the point cloud and improving the accuracy of target detection.

[0108] Assume: in the n-frame point cloud, for the static target G2, since target G2 is a static object relative to the laser radar, the points of target G2 in the first frame point cloud and the target frame point cloud should be the same; when the aligned point cloud of the first frame point cloud is obtained by the scene flow estimation algorithm, the coordinates of the points of target G2 in the aligned point cloud of the first frame point cloud and the coordinates of G2 in the target frame point cloud will have some errors (that is, at the position of target G2, the coordinates of the points in the aligned point cloud are different to a certain extent from the coordinates of the points in the target frame point cloud). Therefore, using the scene flow estimation algorithm for point cloud alignment can increase the density of static objects in the point cloud, thereby achieving point cloud densification and improving target detection accuracy. If the point cloud registration method is used to obtain the aligned point cloud of the first frame point cloud, the points in the aligned point cloud at the position of target G2 are the same as those in the target frame point cloud, that is, point cloud densification cannot be achieved at the position of target G2. Therefore, for static targets, point cloud alignment using scene flow estimation algorithm has better point cloud densification effect than point cloud alignment using point cloud registration method.

[0109] Step S205: perform target detection on the fused frame point cloud to obtain target detection results corresponding to the N frame point cloud.

[0110] In the embodiment, after obtaining the fused frame point cloud, target detection is performed on the fused frame point cloud to obtain a target detection result, which corresponds to the N frame point cloud.

[0111] In some embodiments, target detection is performed on the fused frame point cloud using a target detection algorithm. The target detection algorithm may be any currently available target detection algorithm, which is not described in detail in this application.

[0112] It should be understood that, depending on the different ways of selecting N frames of point clouds, the target detection method in the embodiments of the present application can be applied to different scenarios.

[0113] For example, target detection can be performed in a specific time period, which is frame extraction detection. For example, a portion of N frames of point clouds are extracted from 1000 frames in a specific time period obtained by the lidar to perform target detection.

[0114] For another example, the target can be tracked and detected. In this case, for all point clouds collected by the lidar, each time a frame of point cloud is obtained, target detection is performed on the consecutive N frames of point cloud including the current frame of point cloud.

[0115] In the above target detection method, the point clouds are aligned, and then the aligned point clouds of other frames are fused with the target frame point cloud to obtain the fused frame point cloud, thereby achieving the densification of the point cloud; target detection based on the densified fused frame point cloud can improve the detection rate of target detection and the accuracy of target detection. This method can make sparse point clouds dense at the algorithm level without making changes to the hardware (such as increasing the number of radar lines, etc.), saving costs. In addition, this method can effectively use multi-frame information to improve target detection of sparse point clouds. Even if the radar does not generate a point cloud for the target in a few frames, it can be supplemented by other frame information, so it can be applied to target detection in complex intersection scenes with high robustness.

[0116] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0117] Corresponding to a target detection method in the above embodiment, Figure 5 A structural block diagram of a target detection device provided in an embodiment of the present application is shown. For ease of explanation, only the parts related to the embodiment of the present application are shown.

[0118] Reference Figure 5 , the target detection device 500 includes: an acquisition unit 510, a determination unit 520, an alignment unit 530, a fusion unit 540, and a target detection unit 550, wherein:

[0119] An acquisition unit 510 is used to acquire N consecutive frames of point clouds collected by the laser radar, where N is a positive integer greater than or equal to 2;

[0120] A determination unit 520, configured to determine a frame of point cloud among the N frames of point cloud as a target frame of point cloud;

[0121] The alignment unit 530 is used to align N-1 frames of point clouds other than the target frame point cloud in the N frame point cloud to the target frame point cloud, respectively, to obtain N-1 frames of aligned point clouds, and the position of a point in any frame of the aligned point cloud is the estimated position of the point in the point cloud before alignment at the corresponding time of the target frame point cloud;

[0122] A fusion unit 540 is used to fuse the N-1 frame aligned point cloud with the target frame point cloud to obtain a fused frame point cloud;

[0123] The target detection unit 550 is used to perform target detection on the fused frame point cloud to obtain the target detection result corresponding to the N frame point cloud.

[0124] In one embodiment, the alignment unit 530 is used to align N-1 frames of point clouds other than the target frame point cloud in the N frames of point clouds to the target frame point cloud to obtain the N-1 frames of aligned point clouds, including:

[0125] According to the first frame point cloud and the target frame point cloud, determining position conversion information corresponding to the first frame point cloud, the position conversion information including position change parameters of points in the first frame point cloud converted to estimated positions, the first frame point cloud being any one frame point cloud in the N-1 frame point clouds;

[0126] The points in the first frame point cloud are transformed using the position transformation information to obtain an aligned point cloud corresponding to the first frame point cloud.

[0127] In one embodiment, the position conversion information includes a motion vector, and the alignment unit 530 is used to determine the position conversion information corresponding to the first frame point cloud according to the first frame point cloud and the target frame point cloud, including:

[0128] The first frame point cloud and the target frame point cloud are input into a preset scene flow estimation algorithm to obtain a motion vector corresponding to the first frame point cloud.

[0129] In one embodiment, the motion vector corresponding to the first frame point cloud includes a coordinate transformation vector corresponding to each point in the first frame point cloud;

[0130] The alignment unit 530 is used to transform the points in the first frame point cloud using the position transformation information to obtain an aligned point cloud corresponding to the first frame point cloud, including:

[0131] Add each point in the first frame point cloud to the corresponding coordinate transformation vector to obtain the aligned point cloud corresponding to the first frame point cloud.

[0132] In one embodiment, the position conversion information includes a transformation matrix, and the alignment unit 530 is used to determine the position conversion information corresponding to the first frame point cloud according to the first frame point cloud and the target frame point cloud, including:

[0133] The first frame point cloud and the target frame point cloud are input into a preset point cloud registration algorithm to obtain a transformation matrix corresponding to the first frame point cloud.

[0134] In one embodiment, the data collected by the laser radar includes M consecutive frames of point clouds, and the N frame point cloud is any consecutive N frame point cloud in the M frame point cloud, where M is a positive integer greater than N.

[0135] In one embodiment, the target frame point cloud is the Kth frame point cloud in time sequence among N frame point clouds, where 1≤K≤N and K is a positive integer.

[0136] In one embodiment, the target frame point cloud is a frame point cloud having the largest number of points among N frames point clouds.

[0137] In one embodiment, the laser radar is a roadside laser radar, and the N-frame point clouds are point clouds obtained by the laser radar scanning the same scanning area at different times; the scanning area includes an area of ​​interest and an area of ​​non-interest, the area of ​​interest is the area where the target to be detected appears, the target frame point cloud is a frame point cloud that includes the most target points in the N-frame point clouds, and the target point is a point in the point cloud located in the area of ​​interest.

[0138] Figure 6 A schematic diagram of the structure of a computer device 6 provided in an embodiment of the present application is shown in FIG. Figure 6 As shown, the computer device 6 of this embodiment includes: at least one processor 600 ( Figure 6 Only one processor is shown in the figure), a memory 601, and a computer program 602 stored in the memory 601 and executable on the at least one processor 600, wherein the processor 600 implements the steps in any of the above-mentioned target detection method embodiments when executing the computer program 602.

[0139] The computer device may include, but is not limited to, a processor 600 and a memory 601. Those skilled in the art will appreciate that Figure 6 It is only an example of the computer device 6 and does not constitute a limitation on the computer device 6. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, etc.

[0140] The processor 600 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0141] The memory 601 may be an internal storage unit in some embodiments, such as a hard disk or a memory. The memory 601 may also be an external storage device in other embodiments, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Further, the memory 601 may also include both an internal storage unit and an external storage device. The memory 601 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program, etc. The memory 601 may also be used to temporarily store data that has been output or is to be output.

[0142] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0143] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0144] The embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.

[0145] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0146] Get N consecutive frames of point cloud collected by the laser radar, where N is a positive integer greater than or equal to 2;

[0147] Determine one frame of point cloud among N frames of point cloud as the target frame of point cloud;

[0148] Align the N-1 frames of point clouds except the target frame point cloud in the N frame point cloud to the target frame point cloud, and obtain the N-1 frames of aligned point clouds. The position of a point in any frame of aligned point cloud is the estimated position of the point in the point cloud before alignment at the corresponding time of the target frame point cloud.

[0149] Fuse the aligned point cloud of the N-1 frame with the point cloud of the target frame to obtain the fused frame point cloud;

[0150] Perform target detection on the fused frame point cloud to obtain the target detection results corresponding to the N frame point cloud.

[0151] In one embodiment, the computer program, when executed by a processor, implements the following steps:

[0152] According to the first frame point cloud and the target frame point cloud, determining position conversion information corresponding to the first frame point cloud, the position conversion information including position change parameters of points in the first frame point cloud converted to estimated positions, the first frame point cloud being any one frame point cloud in the N-1 frame point clouds;

[0153] The points in the first frame point cloud are transformed using the position transformation information to obtain an aligned point cloud corresponding to the first frame point cloud.

[0154] In one embodiment, the position conversion information includes a motion vector, and the computer program implements the following steps when executed by a processor: inputting the first frame point cloud and the target frame point cloud into a preset scene flow estimation algorithm to obtain a motion vector corresponding to the first frame point cloud.

[0155] In one embodiment, the motion vector corresponding to the first frame point cloud includes a coordinate transformation vector corresponding to each point in the first frame point cloud; when the computer program is executed by the processor, the following steps are implemented: each point in the first frame point cloud is added to the corresponding coordinate transformation vector to obtain an aligned point cloud corresponding to the first frame point cloud.

[0156] In one embodiment, the position conversion information includes a transformation matrix, and the computer program implements the following steps when executed by a processor: inputting the first frame point cloud and the target frame point cloud into a preset point cloud alignment algorithm to obtain a transformation matrix corresponding to the first frame point cloud.

[0157] In one embodiment, the data collected by the laser radar includes M consecutive frames of point clouds, and the N frame point cloud is any consecutive N frame point cloud in the M frame point cloud, where M is a positive integer greater than N.

[0158] In one embodiment, the target frame point cloud is the Kth frame point cloud in time sequence among N frame point clouds, where 1≤K≤N and K is a positive integer.

[0159] In one embodiment, the target frame point cloud is a frame point cloud having the largest number of points among N frames point clouds.

[0160] In one embodiment, the laser radar is a roadside laser radar, and the N-frame point clouds are point clouds obtained by the laser radar scanning the same scanning area at different times; the scanning area includes an area of ​​interest and an area of ​​non-interest, the area of ​​interest is the area where the target to be detected appears, the target frame point cloud is a frame point cloud that includes the most target points in the N-frame point clouds, and the target point is a point in the point cloud located in the area of ​​interest.

[0161] The implementation principles and technical effects of the various steps implemented when the computer program in this embodiment is executed by the processor are similar to the principles of the above-mentioned laser radar-based detection method and will not be repeated here.

[0162] An embodiment of the present application provides a computer program product. When the computer program product runs on a laser radar, the laser radar can implement the steps in the above-mentioned method embodiments when executed.

[0163] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0164] Get N consecutive frames of point cloud collected by the laser radar, where N is a positive integer greater than or equal to 2;

[0165] Determine one frame of point cloud among N frames of point cloud as the target frame of point cloud;

[0166] Align the N-1 frames of point clouds except the target frame point cloud in the N frame point cloud to the target frame point cloud, and obtain the N-1 frames of aligned point clouds. The position of a point in any frame of aligned point cloud is the estimated position of the point in the point cloud before alignment at the corresponding time of the target frame point cloud.

[0167] Fuse the aligned point cloud of the N-1 frame with the point cloud of the target frame to obtain the fused frame point cloud;

[0168] Perform target detection on the fused frame point cloud to obtain the target detection results corresponding to the N frame point cloud.

[0169] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0170] According to the first frame point cloud and the target frame point cloud, determining position conversion information corresponding to the first frame point cloud, the position conversion information including position change parameters of points in the first frame point cloud converted to estimated positions, the first frame point cloud being any one frame point cloud in the N-1 frame point clouds;

[0171] The points in the first frame point cloud are transformed using the position transformation information to obtain an aligned point cloud corresponding to the first frame point cloud.

[0172] In one embodiment, the position conversion information includes a motion vector, and the computer program, when executed by a processor, further implements the following steps: inputting the first frame point cloud and the target frame point cloud into a preset scene flow estimation algorithm to obtain a motion vector corresponding to the first frame point cloud.

[0173] In one embodiment, the motion vector corresponding to the first frame point cloud includes a coordinate transformation vector corresponding to each point in the first frame point cloud; when the computer program is executed by the processor, the following steps are also implemented: each point in the first frame point cloud is added to the corresponding coordinate transformation vector to obtain an aligned point cloud corresponding to the first frame point cloud.

[0174] In one embodiment, the position conversion information includes a transformation matrix, and the computer program, when executed by a processor, further implements the following steps: inputting the first frame point cloud and the target frame point cloud into a preset point cloud registration algorithm to obtain a transformation matrix corresponding to the first frame point cloud.

[0175] In one embodiment, the data collected by the laser radar includes M consecutive frames of point clouds, and the N frame point cloud is any consecutive N frame point cloud in the M frame point cloud, where M is a positive integer greater than N.

[0176] In one embodiment, the target frame point cloud is the Kth frame point cloud in time sequence among N frame point clouds, where 1≤K≤N and K is a positive integer.

[0177] In one embodiment, the target frame point cloud is a frame point cloud having the largest number of points among N frames point clouds.

[0178] In one embodiment, the laser radar is a roadside laser radar, and the N-frame point clouds are point clouds obtained by the laser radar scanning the same scanning area at different times; the scanning area includes an area of ​​interest and an area of ​​non-interest, the area of ​​interest is the area where the target to be detected appears, the target frame point cloud is a frame point cloud that includes the most target points in the N-frame point clouds, and the target point is a point in the point cloud located in the area of ​​interest.

[0179] The implementation principles and technical effects of the various steps implemented when the computer program in this embodiment is executed by the processor are similar to the principles of the above-mentioned laser radar-based detection method and will not be repeated here.

[0180] An embodiment of the present application further provides a chip, including: a processor, configured to call and run a computer program from a memory, so that a computer device equipped with the chip executes the steps in the above-mentioned method embodiments.

[0181] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0182] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0183] In the embodiments provided in the present application, it should be understood that the disclosed devices / equipment and methods can be implemented in other ways. For example, the device / equipment embodiments described above are merely schematic, for example, the division of the modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0184] 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 on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0185] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0186] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media do not include electric carrier signals and telecommunication signals.

[0187] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A target detection method, characterized in that: The method comprises: Get N consecutive frames of point cloud collected by the laser radar, where N is a positive integer greater than or equal to 2; Determine a frame of point cloud among the N frames of point cloud as a target frame of point cloud; Align N-1 frames of point clouds other than the target frame point cloud in the N frame point cloud to the target frame point cloud, respectively, to obtain N-1 frames of aligned point clouds, the position of a point in any frame of aligned point cloud being the estimated position of the point in the point cloud before alignment at the corresponding time of the target frame point cloud; Fusing the N-1 frame aligned point cloud with the target frame point cloud to obtain a fused frame point cloud; Perform target detection on the fused frame point cloud to obtain target detection results corresponding to the N frame point clouds.

2. The method according to claim 1, characterized in that The step of aligning N-1 frames of point clouds other than the target frame point cloud in the N frames of point clouds to the target frame point cloud to obtain N-1 frames of aligned point clouds includes: Determine, according to the first frame point cloud and the target frame point cloud, position conversion information corresponding to the first frame point cloud, the position conversion information including position change parameters of points in the first frame point cloud converted to the estimated positions, the first frame point cloud being any one of the N-1 frame point clouds; The points in the first frame point cloud are transformed using the position transformation information to obtain an aligned point cloud corresponding to the first frame point cloud.

3. The method according to claim 2, characterized in that The position conversion information includes a motion vector, and determining the position conversion information corresponding to the first frame point cloud according to the first frame point cloud and the target frame point cloud includes: The first frame point cloud and the target frame point cloud are input into a preset scene flow estimation algorithm to obtain a motion vector corresponding to the first frame point cloud.

4. The method according to claim 3, characterized in that The motion vector corresponding to the first frame point cloud includes a coordinate conversion vector corresponding to each point in the first frame point cloud; The step of using the position conversion information to convert points in the first frame point cloud to obtain an aligned point cloud corresponding to the first frame point cloud includes: Each point in the first frame point cloud is added to the corresponding coordinate transformation vector to obtain an aligned point cloud corresponding to the first frame point cloud.

5. The method according to claim 2, characterized in that: The position conversion information includes a transformation matrix, and determining the position conversion information corresponding to the first frame point cloud according to the first frame point cloud and the target frame point cloud includes: The first frame point cloud and the target frame point cloud are input into a preset point cloud registration algorithm to obtain the transformation matrix corresponding to the first frame point cloud.

6. The method according to any one of claims 1 to 5, characterized in that: The data collected by the laser radar includes M continuous frames of point clouds, and the N frame point cloud is any continuous N frame point cloud in the M frame point cloud, where M is a positive integer greater than N.

7. The method according to claim 6, characterized in that The target frame point cloud is the Kth frame point cloud in time sequence among the N frame point clouds, where 1≤K≤N and K is a positive integer.

8. The method according to any one of claims 1 to 6, characterized in that: The target frame point cloud is a frame point cloud having the largest number of points among the N frames point clouds.

9. The method according to any one of claims 1 to 6, characterized in that: The laser radar is a roadside laser radar, and the N frames of point clouds are point clouds obtained by the laser radar scanning the same scanning area at different times; the scanning area includes an area of ​​interest and an area of ​​non-interest, and the area of ​​interest is the area where the target to be detected appears. The target frame point cloud is a frame of point cloud that includes the most target points in the N frames of point clouds, and the target point is a point in the point cloud located in the area of ​​interest.

10. A target detection device, characterized in that: The device comprises means for performing the steps of the method according to any one of claims 1 to 9.

11. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 9 is implemented.

12. A target detection system, characterized in that: Comprising a laser radar and the computer device described in claim 11.

13. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.

14. A chip, characterized in that: include: A processor, configured to call and run a computer program from a memory, so that a computer device equipped with the chip executes a method as claimed in any one of claims 1 to 9.

Images