Target recognition method and device based on laser radar speed measurement

Abstract

The application provides a target identification method and device based on laser radar speed measurement, and the method comprises the following steps: extracting adjacent two frame point cloud images from acquired laser radar point cloud data for splicing to obtain spliced point cloud data; inputting the spliced point cloud data into a target identification model to obtain a target identification result output by the target identification model; wherein the target identification model is trained based on point cloud training data and corresponding target identification true value. The application identifies the target through the target identification model for the spliced point cloud data after splicing, so as to improve the target identification precision, avoid the mis-detection or mis-association of a static target crossing a road, further ensure the safety of vehicle automatic driving, and reduce the probability of accidents.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic driving, and in particular to a target identification method and device based on laser radar speed measurement. BACKGROUND

[0002] An autonomous vehicle, also known as a self-piloting automobile, is an intelligent automobile that realizes unmanned driving through a computer system. Autonomous driving includes an assisted driving vehicle that assists a driver to drive and a fully automated driving unmanned vehicle. With the continuous deepening of research on autonomous driving at home and abroad at the present stage, autonomous driving technology gradually promotes from assisted driving to unmanned driving. Regardless of which stage of autonomous driving research, the performance of the vehicle needs to be detected to prove or improve the safety of the vehicle. Among them, the speed measurement of the front target in the driving process is particularly important.

[0003] The front target is generally a front driving vehicle of the current vehicle. In existing assisted or autonomous driving, the speed measurement of the front target is mostly completed by a millimeter wave radar. When the target approaches the millimeter wave radar antenna, the reflected signal frequency will be higher than the transmitter frequency; on the contrary, when the target moves away from the antenna, the reflected signal frequency will be lower than the transmitter frequency. In this way, the relative speed of the target and the millimeter wave radar can be calculated by changing the value of the frequency.

[0004] However, the millimeter wave radar speed measurement cannot obtain the specific category of the target. When the ego vehicle drives on the highway, the millimeter wave radar will identify the front gantry, overpass, and interchange as a static target across the road, resulting in the ego vehicle braking in the autonomous driving state. SUMMARY

[0005] The present application provides a target identification method and device based on laser radar speed measurement to solve the defect that the millimeter wave radar cannot obtain the specific category of the target when measuring speed in the prior art, realize the identification of the target category while measuring the speed of the target, and improve the target detection performance.

[0006] The present application provides a target identification method based on laser radar speed measurement, comprising: extracting adjacent two frames of point cloud images from the obtained laser radar point cloud data for splicing to obtain spliced point cloud data; inputting the spliced point cloud data into a target identification model to obtain a target identification result output by the target identification model; wherein the target identification model is trained based on point cloud training data and corresponding target identification true value.

[0007] The application provides a target identification method based on laser radar speed measurement.

[0008] The application provides a target identification method based on laser radar speed measurement.

[0009] The application provides a target identification method based on laser radar speed measurement.

[0010] The application provides a target identification method based on laser radar speed measurement.

[0011] The application provides a target identification method based on laser radar speed measurement.

[0012] The application further provides a target identification device based on laser radar speed measurement, comprising: a splicing module which extracts adjacent two frames of point cloud images from acquired point cloud data for splicing to obtain spliced point cloud data; the point cloud data is obtained by detecting a front road of a vehicle based on a laser radar; an identification module which inputs the spliced point cloud data into a target identification model to obtain a target identification result output by the target identification model; wherein the target identification model is trained based on point cloud training data and corresponding target identification true values.

[0013] The application further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the target identification method based on laser radar speed measurement according to any one of the above when executing the program.

[0014] The application further provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the steps of the target identification method based on laser radar speed measurement according to any one of the above.

[0015] The application further provides a computer program product comprising a computer program, wherein the computer program is executable by a processor to implement the steps of the target identification method based on laser radar speed measurement according to any one of the above.

[0016] The target identification method and device based on laser radar speed measurement provided by the application splice point cloud images of adjacent two frames in laser radar point cloud data to make the point cloud denser, thereby facilitating improvement of the identification accuracy of a subsequent target identification model; the target identification model is used to identify a target based on spliced point cloud data obtained by splicing laser radar point cloud data, to obtain a target identification result, improve target detection performance, further avoid false detection or false association of a static target such as a gantry across a road by radar speed measurement, ensure the safety of vehicle automatic driving, and reduce the probability of accidents. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and those skilled in the art can also obtain other drawings according to these drawings without creative effort.

[0018] Figure 1 is a flowchart of the target identification method based on laser radar speed measurement provided by the application;

[0019] Figure 2 is a training flowchart of the target identification model provided by the application;

[0020] Figure 3 is a structural schematic diagram of a target recognition device based on laser radar speed measurement provided by the application;

[0021] Figure 4 is a structural schematic diagram of a training module provided by the application;

[0022] Figure 5 is a structural schematic diagram of an electronic device provided by the application. DETAILED DESCRIPTION

[0023] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0024] Figure 1 A flowchart of a target recognition method based on laser radar speed measurement is shown, and the method comprises the following steps:

[0025] S01, adjacent two frames of point cloud images are extracted from the obtained point cloud data for splicing to obtain spliced point cloud data; the point cloud data is obtained by detecting the road in front of the vehicle based on the laser radar.

[0026] S02, the spliced point cloud data is input into a target recognition model to obtain a target recognition result output by the target recognition model; wherein the target recognition model is trained based on point cloud training data and corresponding target recognition true value.

[0027] It should be noted that S0N in the present specification does not represent the order of the target recognition method based on laser radar speed measurement, and the following will be specifically combined with Figure 2 The target recognition method based on laser radar speed measurement of the present application is described.

[0028] Step S01, adjacent two frames of point cloud images are extracted from the obtained point cloud data for splicing to obtain spliced point cloud data; the point cloud data is obtained by detecting the road in front of the vehicle based on the laser radar.

[0029] In the embodiment, the adjacent two frame point cloud images are extracted from the acquired point cloud data for splicing, including: extracting the adjacent two frame point cloud images from the acquired point cloud data; performing coordinate conversion on the previous frame point cloud image in the adjacent two frame point cloud images based on the next frame point cloud image; and splicing the coordinate-converted previous frame point cloud image to the next frame point cloud image. It should be noted that the spliced point cloud image includes information of each point in the corresponding point cloud data, specifically including position information of each point and speed information of each point.

[0030] Specifically, the coordinate conversion of the previous frame point cloud in the adjacent two frame point cloud based on the next frame point cloud image includes: performing coordinate conversion on the previous frame point cloud image in the adjacent two frame point cloud images based on the world coordinate system to obtain intermediate conversion data; and performing coordinate conversion on the intermediate conversion data based on the next frame point cloud image in the adjacent two frame point cloud images. Further, the coordinate conversion of the intermediate conversion data based on the next frame point cloud image in the adjacent two frame point cloud images includes: performing coordinate conversion on the next frame point cloud image based on the world coordinate system to obtain reference coordinate data; and converting the intermediate conversion data to the coordinate system of the next frame point cloud image based on the reference coordinate data.

[0031] It should be noted that through the world coordinate system, the previous frame point cloud image is first converted from its coordinate system to the world coordinate system, and then based on the relationship between the coordinate system of the next frame point cloud image and the world coordinate system, the previous frame point cloud image is converted from the world coordinate system to the coordinate system of the next frame point cloud image, thereby facilitating subsequent splicing of the previous frame point cloud image to the next frame point cloud image.

[0032] In an optional embodiment, before splicing the coordinate-converted previous frame point cloud image to the next frame point cloud image, it further includes: adding a first feature as a time feature to the previous frame point cloud image; and adding a second feature as a time feature to the next frame point cloud image.

[0033] It should be noted that the first feature and the second feature are used as time features to distinguish the point cloud images belonging to the previous frame and the point cloud images belonging to the next frame in the spliced point cloud data, thereby facilitating subsequent learning of speed information of the corresponding target based on the time features. In addition, by adding the time features, it is convenient for subsequent learning of the speed information of the target by using the target recognition model, and then obtaining the target speed.

[0034] In an optional embodiment, before the adjacent two frames of point cloud data are extracted from the acquired point cloud data for splicing, the following is included: detecting a road ahead of the vehicle based on the laser radar to obtain the point cloud data. It should be noted that the point cloud data, i.e., three-dimensional laser radar point cloud data, is data acquired by scanning the road ahead of the vehicle by the laser radar; the laser radar is an active remote sensing device using a laser as a light source and using an optoelectronic detection technology, mainly composed of a transmitting system, a receiving system, and an information processing system, which is not limited in this embodiment.

[0035] In addition, the automatic driving vehicle can be provided with a laser radar device (such as front and rear laser radar devices), a camera device (such as a front camera), and a sensing device (such as a left rear wheel sensor), etc., which are not limited in this embodiment. Among them, the laser radar device can scan the surrounding environment within a certain radius and present the results in the form of a 3D map, providing the computer device with the initial basis for judgment. The front and rear laser radar devices can also be combined with the front camera to measure the distance between the automatic driving vehicle and each object in front, back, left, and right.

[0036] It should be noted that, in order to improve the accuracy of subsequent model prediction, after acquiring the point cloud data, the point cloud data can also be preprocessed to effectively filter out the clutter interference of individual radar points. Specifically, the median filter is used to smooth the point cloud data and remove isolated noise points. In other embodiments, other preprocessing methods can also be used to remove the clutter interference of individual radar points.

[0037] In an optional embodiment, after acquiring the point cloud data, the following is also included: clustering the point cloud data, so that the point cloud data corresponding to similar targets are clustered into the same cluster, so that the subsequent target recognition model can recognize each cluster data in the clustering result, thereby reducing the computational load of the subsequent target recognition model.

[0038] Step S02: inputting the spliced point cloud data into the target recognition model to obtain a target information recognition result output by the target recognition model; wherein the target recognition model is trained based on point cloud training data and its corresponding target recognition true value.

[0039] In this embodiment, the spliced point cloud data includes a plurality of spliced point cloud images, and the target recognition model is used for target recognition based on the extracted speed features and position features of the spliced point cloud data to output a target recognition result of the corresponding target, which includes target category, target speed, target position, and target size, etc. information, so as to control the automatic driving of the vehicle.

[0040] It should be noted that when the target recognition model identifies the spliced point cloud data, since the spliced point cloud data contains position information in the height direction perpendicular to the ground surface, the target recognition model can determine the specific category of the target according to the information, such as a vehicle, a pedestrian, a stationary target located on a driving road, and a gantry, a bridge, an overpass, and the like, which are identified as a stationary target across the road, thereby avoiding subsequent mis-detection of the gantry and the like, stationary target categories across the road, avoiding subsequent automatic control of the vehicle according to the mis-detected target, and further ensuring the safety of the vehicle automatic driving and reducing the probability of accidents.

[0041] Reference Figure 2 In an optional embodiment, the target recognition method based on laser radar speed measurement further comprises training the target recognition model, specifically comprising:

[0042] S11, acquiring point cloud training data;

[0043] S12, obtaining target recognition true value according to the point cloud training data;

[0044] S13, extracting adjacent two frames of point cloud from the point cloud training data for splicing, and inputting the spliced point cloud training data into the target recognition model to obtain a target training result;

[0045] S14, constructing a loss function based on the target training result and the target recognition true value, and ending the training based on the convergence of the loss function.

[0046] Specifically:

[0047] First, the point cloud training data is acquired. In this embodiment, the acquired point cloud training data is point cloud data for training. The jth point in the ith frame of point cloud training image is represented as (x ij ,y ij ,z ij ).

[0048] Second, the target recognition true value is obtained according to the point cloud training data. In this embodiment, the target recognition true value includes a category true value corresponding to a target category, a speed true value corresponding to a target speed, a position true value corresponding to a target position, and a size true value corresponding to a target size. For example, when the target recognition true value includes the speed true value, the target recognition true value is obtained according to the point cloud training data, including: obtaining the motion trajectory of the target according to the point cloud training data; calculating the speed of the target in each frame of point cloud training data according to the motion trajectory to obtain the speed true value. It should be noted that the calculation formula of the speed true value at time t is represented as:

[0049]

[0050]

[0051] vx(t) = vx(t) + vx(t) * Δt t vx(t) = vx(t) + vx(t) * Δt t vy(t) = vy(t) + vy(t) * Δt next Δt = t + Δt prev Δt = t - Δt next t = t + Δt prev Δt = t - Δt t+Δtnext x(t) = x(t) + vx(t) * Δt t+Δtprev x(t) = x(t) - vx(t) * Δt t+Δtnext y(t) = y(t) + vy(t) * Δt t+Δtprev y(t) = y(t) - vy(t) * Δt

[0052] It should be noted that the front direction of the vehicle body can be taken as the positive direction of the x-axis, the left direction of the vehicle body is perpendicular to the ground plane and the positive direction of the y-axis, and the upward direction perpendicular to the ground plane is the positive direction of the z-axis.

[0053] Subsequently, the adjacent two frames of point clouds are extracted from the point cloud training data for splicing, and the spliced point cloud training data is input into the target recognition model to obtain a target training result. In this embodiment, the target training result includes target training categories, target training speeds, target training positions, target training sizes and the like.

[0054] Finally, based on the target training result and the target recognition true value, a loss function is constructed, and the training is ended based on the convergence of the loss function. Taking the target training speed output by the model as an example, in order to improve the accuracy of the target vehicle speed prediction, the calculation method of the corresponding loss function L is as follows:

[0055]

[0056] L = 1 / N * åi=1N (vx(t) - vxtrue(t))2+ (vy(t) - vytrue(t))2 i vx(t) = vx(t) + vx(t) * Δt vxtrue(t) = vxtrue(t) + vxtrue(t) * Δt i vy(t) = vy(t) + vy(t) * Δt vytrue(t) = vytrue(t) + vytrue(t) * Δt

[0057] It should be noted that the smaller the loss function is, the higher the accuracy of the target recognition model is. When the calculated loss function tends to converge, the training ends. In addition, the target training class, the target training position and the target training size of the model output can be calculated according to the class true value, the position true value and the size true value respectively according to the above loss function construction manner, so as to improve the prediction accuracy of the model for the target class, the target position and the target size.

[0058] In summary, the embodiment of the present application splices the point cloud images of the adjacent two frames in the laser radar point cloud data, so that the point cloud is more dense, thereby facilitating to improve the recognition accuracy of the subsequent target recognition model; the target recognition model is used to perform target recognition on the point cloud splicing data spliced based on the laser radar point cloud data, to obtain a target recognition result, improve the target detection performance, further avoid the false detection or false association of the radar speed on the static target such as the gantry crossing the road, ensure the safety of the vehicle automatic driving, and reduce the probability of accidents.

[0059] The target recognition device based on laser radar speed provided by the present application is described below, and the target recognition device based on laser radar speed described below can be correspondingly referred to the target recognition method based on laser radar speed described above.

[0060] Figure 3 The structure of the target recognition device based on laser radar speed is shown, and the device comprises:

[0061] The splicing module 31 extracts the adjacent two frames of point cloud images from the obtained point cloud data for splicing, to obtain spliced point cloud data; the point cloud data is obtained by detecting the front road of the vehicle based on the laser radar;

[0062] The recognition module 32 inputs the spliced point cloud data into the target recognition model to obtain the target recognition result output by the target recognition model; wherein the target recognition model is trained based on the point cloud training data and the corresponding target recognition true value.

[0063] In the embodiment, the splicing module 31 comprises: a point cloud extraction unit for extracting the adjacent two frames of point cloud images from the obtained point cloud data; a coordinate conversion unit for converting the coordinates of the previous frame of point cloud images in the adjacent two frames of point cloud images based on the coordinates of the next frame of point cloud images; and a splicing unit for splicing the previous frame of point cloud images after coordinate conversion to the next frame of point cloud images. It should be noted that the spliced point cloud image comprises the information of each point in the corresponding point cloud data, specifically comprising the position information of each point and the speed information of each point.

[0064] Further, the coordinate conversion unit comprises: a first conversion sub-unit configured to perform coordinate conversion on a previous frame of point cloud images in the two adjacent frames of point cloud images based on the world coordinate system to obtain intermediate conversion data; and a second conversion sub-unit configured to perform coordinate conversion on the intermediate conversion data based on a subsequent frame of point cloud images in the two adjacent frames of point cloud images. Further, the second conversion sub-unit comprises: a reference coordinate conversion sub-sub-unit configured to perform coordinate conversion on the subsequent frame of point cloud images based on the world coordinate system to obtain reference coordinate data; and a point cloud coordinate conversion sub-sub-unit configured to convert the intermediate conversion data to the coordinate system in which the subsequent frame of point cloud images is located based on the reference coordinate data.

[0065] In an optional embodiment, the splicing module 31 further comprises: a first feature adding unit configured to add a first feature as a time feature to the previous frame of point cloud images; and a second feature adding unit configured to add a second feature as a time feature to the subsequent frame of point cloud images. It should be noted that the first feature and the second feature are used as time features to distinguish the point cloud images belonging to the previous frame and the point cloud images belonging to the subsequent frame in the spliced point cloud data, so as to facilitate subsequent learning of speed information of the corresponding target based on the time features. In addition, the time features are added to facilitate subsequent learning of speed information of the target by using the target recognition model, and then the speed of the target is obtained.

[0066] In an optional embodiment, the device further comprises: a data acquisition module configured to detect a road ahead of the vehicle based on the lidar to obtain the point cloud data. It should be noted that the point cloud data, i.e., three-dimensional lidar point cloud data, is data obtained by scanning the road ahead of the vehicle by the lidar; and the lidar is an active remote sensing device using a laser as a light source and adopting an optoelectronic detection technology, mainly composed of a transmitting system, a receiving system, and an information processing system, which are not limited in this embodiment.

[0067] In an optional embodiment, the device further comprises: a data preprocessing module configured to preprocess the point cloud data to effectively filter out the clutter interference of a single radar point. Specifically, the data preprocessing module comprises: a smoothing processing unit configured to perform smoothing processing on the point cloud data by using median filtering to remove isolated noise points.

[0068] In an optional embodiment, the device further comprises: a clustering module configured to cluster the point cloud data, so that the point cloud data corresponding to similar targets are clustered into the same cluster, so as to facilitate subsequent identification of the target recognition model for each cluster data in the clustering result, thereby reducing the calculation amount of the subsequent target recognition model.

[0069] The recognition module 32 comprises: an input unit configured to input the spliced point cloud data to a target recognition model unit; the target recognition model unit configured to recognize the input spliced point cloud data to obtain a target recognition result; and an output unit configured to output the target recognition result. In this embodiment, the spliced point cloud data comprises a plurality of spliced point cloud images, and the target recognition model is configured to recognize a target based on the extracted speed feature and position feature of the spliced point cloud data to output a recognition result of the corresponding target, and the target recognition result comprises information such as a target category, a target speed, a target position and a target size, so as to facilitate automatic driving of the vehicle.

[0070] It should be noted that, when the target recognition model recognizes the spliced point cloud data, since the spliced point cloud data contains position information in the height direction perpendicular to the ground plane, the target recognition model can determine the specific category of the target according to the information, such as a vehicle, a pedestrian, a stationary target on a driving road, and a gantry, a bridge, an overpass, etc. recognized as a stationary target across the road, thereby avoiding subsequent false detection of the stationary target across the road such as the gantry, and avoiding subsequent control of the automatic driving of the vehicle based on the false detection target, thereby further ensuring the safety of the automatic driving of the vehicle and reducing the probability of accidents.

[0071] In an optional embodiment, the device further comprises a training module 33 configured to train the target recognition model unit.

[0072] Reference Figure 4 The training module comprises: a data acquisition unit 41 configured to acquire point cloud training data; a true value acquisition unit 42 configured to obtain a target recognition true value based on the point cloud training data; a training unit 43 configured to splice adjacent two frames of point cloud from the point cloud training data, and input the spliced point cloud training data to the target recognition model to obtain a target training result; and a loss function acquisition unit 44 configured to construct a loss function based on the target training result and the target recognition true value, and end the training based on convergence of the loss function.

[0073] Specifically, the true value acquisition unit 42 comprises: a trajectory acquisition subunit configured to obtain a motion trajectory of a target based on the point cloud training data; and a true value acquisition subunit configured to calculate a speed of the target in each frame of point cloud training data based on the motion trajectory to obtain a speed true value.

[0074] Figure 5 An example of a schematic diagram of the physical structure of an electronic device is shown in FIG. 1. Figure 5As shown, the electronic device can include a processor 51, a communications interface 52, a memory 53, and a communications bus 54, wherein the processor 51, the communications interface 52, and the memory 53 communicate with each other through the communications bus 54. The processor 51 can invoke the logical instructions in the memory 53 to execute the target identification method based on laser radar speed measurement, which includes: extracting adjacent two frames of point cloud images from the obtained point cloud data for splicing to obtain spliced point cloud data; the point cloud data is obtained based on laser radar detection of the vehicle's forward road; inputting the spliced point cloud data into a target identification model to obtain a target identification result output by the target identification model; wherein the target identification model is trained based on point cloud training data and its corresponding target identification true value.

[0075] In addition, the logical instructions in the memory 53 described above can be implemented in the form of a software function unit and sold or used as an independent product, which can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0076] On the other hand, the present application also provides a computer program product, which includes a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to execute the target identification method based on laser radar speed measurement provided by the above-mentioned methods, which includes: extracting adjacent two frames of point cloud images from the obtained point cloud data for splicing to obtain spliced point cloud data; the point cloud data is obtained based on laser radar detection of the vehicle's forward road; inputting the spliced point cloud data into a target identification model to obtain a target identification result output by the target identification model; wherein the target identification model is trained based on point cloud training data and its corresponding target identification true value.

[0077] In yet another aspect, the present application also provides a non-transitory computer readable storage medium having stored thereon a computer program, which, when executed by a processor, implements a target identification method based on laser radar speed measurement provided by each of the above methods, the method comprising: stitching adjacent two frame point cloud images extracted from acquired point cloud data to obtain stitched point cloud data; the point cloud data being obtained based on laser radar detection of a front road of a vehicle; inputting the stitched point cloud data into a target identification model to obtain a target identification result output by the target identification model; wherein the target identification model is trained based on point cloud training data and corresponding target identification true value.

[0078] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement it without creative labor.

[0079] From the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software and necessary general hardware platform, and of course, it can also be realized by hardware. Based on such understanding, the above technical solutions, essentially or in the form of software products, can be embodied in a computer software product, which can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, server, or network device, etc.) execute the methods described in each embodiment or some parts of the embodiment.

[0080] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A target recognition method based on lidar velocity measurement, characterized in that, include: The point cloud data is obtained by extracting two adjacent frames of point cloud images from the acquired point cloud data and stitching them together; the point cloud data is obtained by detecting the road ahead of the vehicle based on lidar. The stitched point cloud data is input into the target recognition model to obtain the target recognition result output by the target recognition model; wherein, the target recognition model is trained based on the point cloud training data and its corresponding target recognition ground truth values; The target recognition model is used to perform target recognition based on the speed and position features extracted from the stitched point cloud data, so as to output the target recognition result of the corresponding target. The target recognition result includes target category, target speed, target position and target size, so as to control the autonomous driving of the vehicle. The step of extracting and stitching two adjacent frames of point cloud images from the acquired lidar point cloud data includes: Extract point cloud images from two adjacent frames of the acquired lidar point cloud data; The coordinate transformation of the previous point cloud image in the two adjacent point cloud images is based on the next point cloud image. The previous frame point cloud image after coordinate transformation is stitched to the next frame point cloud image; Before stitching the previous frame point cloud image (after coordinate transformation) to the next frame point cloud image, the method further includes: Add a first feature as a temporal feature to the previous frame point cloud image; Add a second feature as a temporal feature to the subsequent frame of the point cloud image; The first feature and the second feature are used as temporal features to distinguish point cloud images belonging to the previous frame and point cloud images belonging to the next frame in the stitched point cloud data.

2. The target recognition method based on lidar velocity measurement according to claim 1, characterized in that, The step of transforming the coordinates of the previous frame point cloud image in two adjacent frame point cloud images based on the next frame point cloud image includes: Based on the world coordinate system, coordinate transformation is performed on the previous frame point cloud image in the two adjacent frame point cloud images to obtain intermediate transformed data; The intermediate transformation data is transformed based on the coordinate transformation of the next point cloud image in the two adjacent point cloud images.

3. The target recognition method based on lidar velocity measurement according to claim 1, characterized in that, Training the target recognition model includes: Acquire point cloud training data; obtain target recognition ground truth based on the point cloud training data; The point cloud training data is extracted from two adjacent frames of point cloud and stitched together. The stitched point cloud training data is then input into the target recognition model to obtain the target training result. Based on the target training results and the target recognition ground truth, a loss function is constructed, and training ends when the loss function converges.

4. The target recognition method based on lidar velocity measurement according to claim 1, characterized in that, The ground truth for target recognition includes a velocity ground truth, and obtaining the ground truth for target recognition based on the point cloud training data includes: Based on the point cloud training data, the target's motion trajectory is obtained; Based on the motion trajectory, the velocity of the target in each frame of point cloud training data is calculated to obtain the true velocity value.

5. A target recognition device based on lidar speed measurement, characterized in that, include: The stitching module extracts two adjacent point cloud images from the acquired point cloud data and stitches them together to obtain stitched point cloud data. The point cloud data is obtained by detecting the road ahead of the vehicle using lidar; The recognition module inputs the stitched point cloud data into the target recognition model to obtain the target recognition result output by the target recognition model; wherein, the target recognition model is trained based on the point cloud training data and its corresponding target recognition ground truth values; The target recognition model is used to perform target recognition based on the speed and position features extracted from the stitched point cloud data, so as to output the target recognition result of the corresponding target. The target recognition result includes target category, target speed, target position and target size, so as to control the autonomous driving of the vehicle. The splicing module includes: The point cloud extraction unit extracts two adjacent frame point cloud images from the acquired lidar point cloud data; The coordinate transformation unit performs coordinate transformation on the previous frame point cloud image in the two adjacent frame point cloud images based on the next frame point cloud image; The stitching unit stitches the previous frame point cloud image after the coordinate transformation to the next frame point cloud image; The splicing module also includes: The first feature adding unit adds a first feature as a time feature to the previous frame point cloud image before stitching the previous frame point cloud image after the coordinate transformation to the next frame point cloud image. The second feature adding unit adds a second feature as a time feature to the subsequent frame point cloud image; The first feature and the second feature are used as temporal features to distinguish point cloud images belonging to the previous frame and point cloud images belonging to the next frame in the stitched point cloud data.

6. An electronic 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 program, it implements the steps of the target recognition method based on lidar speed measurement as described in any one of claims 1 to 4.

7. 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 recognition method based on lidar speed measurement as described in any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the target recognition method based on lidar speed measurement as described in any one of claims 1 to 4.

Images