Vehicle drivable area target recognition method and system based on three-dimensional point cloud

Abstract

The application discloses a vehicle drivable area target recognition method and system based on a three-dimensional point cloud, and comprises the following steps: converting the data of a laser radar on a vehicle, converting the angle and distance information in polar coordinates into coordinates in a Cartesian coordinate system; cutting the converted three-dimensional point cloud, ground segmenting the cut three-dimensional point cloud, obtaining ground point cloud and non-ground point cloud, and taking the ground point cloud as a drivable area; desampling the non-ground point cloud, collecting data and labeling tags, taking the desampled non-ground point cloud as input, the labeled tags as reference, and the detection result as output, and training a neural network model through deep learning; using the trained neural network model to detect and recognize targets, extracting the targets of interest in the drivable area, and publishing target information. The application detects and recognizes obstacles in the drivable area, outputs target information to a decision layer, and provides a reference for a next layer of decision algorithm.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of intelligent vehicles and relates to a vehicle drivable area target recognition method and system based on three-dimensional point clouds. BACKGROUND

[0002] In recent years, intelligentization is one of the key technologies in the automotive industry. Compared with traditional power transmission technology, intelligentization still has a lot of room for development in both hardware and software, and has broad prospects. At present, major automakers have invested a lot of effort and cost in the intelligent technology of vehicles. If the work environment of vehicles with limitations or even certain dangers can be liberated by improving intelligentization, it will be very meaningful.

[0003] The ultimate goal of automotive intelligentization is to achieve high-level automatic driving, and the basis is accurate perception of road information. Common vehicle-mounted perception sensors include cameras, millimeter wave radars and ultrasonic radars, and the development has been relatively mature. In recent years, laser radars have gradually attracted the attention of automakers.

[0004] For three-dimensional point cloud processing algorithms, the research on target segmentation is relatively extensive. Since the characteristics of three-dimensional point clouds are not significant, such as the point clouds reflected on the front and side of large vehicles and the point clouds reflected on the wall, there is almost no difference in characteristics. Therefore, the target recognition algorithm based on three-dimensional point clouds needs further research. SUMMARY

[0005] The purpose of the application is to solve the problems in the prior art and provide a vehicle drivable area target recognition method and system based on three-dimensional point clouds, which detects and recognizes obstacles in the vehicle drivable area through a three-dimensional laser radar.

[0006] To achieve the above purpose, the application adopts the following technical solutions:

[0007] The vehicle drivable area target recognition method based on three-dimensional point clouds comprises the following steps:

[0008] The data of the laser radar on the vehicle is converted to convert the angle and distance information in polar coordinates into coordinates in the Cartesian coordinate system;

[0009] The converted three-dimensional point cloud is cropped, and the cropped three-dimensional point cloud is ground segmented to obtain ground point clouds and non-ground point clouds, and the ground point clouds are used as the drivable area;

[0010] The non-ground point cloud is down-sampled, data is collected and labeled, the down-sampled non-ground point cloud is used as input, the labeled label is used as reference, and the detection result is used as output, and the neural network model is trained through deep learning;

[0011] The trained neural network model is used for target detection and recognition, and the feasible area is extracted for the target of interest, and the target information is published.

[0012] Further, the conversion relationship of the angle and distance information in the polar coordinate to the coordinates in the Cartesian coordinate system is

[0013]

[0014] Wherein, r is the measured distance, ω is the vertical angle of the laser, α is the horizontal rotation angle of the laser, x, y and z are the coordinates of the polar coordinate projection on the X, Y and Z axes of the Cartesian coordinate system.

[0015] Further, the clipping includes clipping the points reflected on the vehicle body and all points higher than the height of the vehicle in the converted three-dimensional point cloud.

[0016] Further, the ground segmentation process is:

[0017] The laser radar emits a cluster of rays, reduces the three-dimensional point cloud (x, y, z) to two-dimensional (x, y), calculates the plane angle θ of each point to the direction in front of the vehicle head, differentiates 360°, and each part has an angle of 0.18°. Each part can be regarded as a ray. The points in these rays are sorted according to the distance.

[0018] Two slope thresholds are set, one is the slope threshold A of the adjacent two points on the same ray, and the other is the slope threshold B of the entire ground. The height threshold is obtained by the two thresholds and the horizontal distance of the current point to the radar. Whether the current point is the ground is determined by judging whether the z value of the current point is within the height threshold range of the ground.

[0019] Further, the downsampling is to segment the non-ground point cloud with a self-defined size cube, and use the center of the cube to represent all point clouds in the cube.

[0020] Further, the process of data collection and label annotation is:

[0021] Three-dimensional point cloud data is collected by recording ROS topics through rosbag. The rosbag is divided into single-frame point clouds in bin format according to the sampling frequency.

[0022] Each frame of point cloud cut out is labeled by LabelColud, and the label annotation includes commercial vehicles, passenger cars, motorcycles, electric vehicles, bicycles, pedestrians and roadblocks.

[0023] Further, the neural network model converts the three-dimensional point cloud into a two-dimensional pseudo image in the form of Pillar, thereby converting 3D convolution into 2D convolution, and the specific implementation steps are as follows:

[0024] The point cloud data is divided into a grid according to the X and Y axes of the point cloud data, and the point cloud data falling into a grid is regarded as being in a Pillar;

[0025] Suppose there are P non-empty Pillars in each sample, and there are N point cloud data in each Pillar, then the sample represents a (D, P, N) tensor, if the point cloud data in each Pillar exceeds N, then it is randomly sampled to N, if the point cloud data in each Pillar is less than N, then the part less than N is filled with 0, and the point cloud is converted into a (D, P, N) tensor format of stacked Pillars;

[0026] After obtaining the stacked Pillars, the tensorized point cloud data is processed and feature extracted, a pooling operation is performed according to the dimension of the Pillar, a (C, P) dimensional feature map is obtained, P is converted into (W, H), and finally a pseudo image in the form of (C, H, W) is obtained.

[0027] A vehicle drivable area target recognition system based on three-dimensional point cloud, comprising:

[0028] A conversion module, configured to convert the data of the laser radar on the vehicle, and convert the angle and distance information in polar coordinates into coordinates in the Cartesian coordinate system;

[0029] A segmentation module, configured to crop the converted three-dimensional point cloud, and perform ground segmentation on the cropped three-dimensional point cloud to obtain ground point cloud and non-ground point cloud, and take the ground point cloud as the drivable area;

[0030] A downsampling module, configured to downsample the non-ground point cloud, collect data and label the label, take the downsampled non-ground point cloud as input, take the label as reference, and take the detection result as output, and train the neural network model through deep learning;

[0031] An extraction module, configured to perform target detection and recognition by using the trained neural network model, extract the target of interest from the drivable area, and publish the target information.

[0032] A terminal device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the method when executing the computer program.

[0033] A computer readable storage medium stores a computer program, the computer program is executed by a processor to realize the steps of the method.

[0034] Compared with the prior art, the present application has the following beneficial effects:

[0035] The present application provides a kind of based on three-dimensional point cloud vehicle drivable area target identification method and system, by the three-dimensional point cloud of laser radar carried on vehicle emission is cut, to the three-dimensional point cloud ground segmentation after cutting, the drivable area of car is segmented, and non-ground point cloud is reduced sampling, the obstacle in drivable area is detected and identified, the position, speed, attitude, length, width, height, category etc. Information of target is output to decision layer, for the next layer decision algorithm reference.

[0036] Further, the present application uses laser radar as vehicle-mounted perception sensor, uses laser detection position, shape and speed and other geometric features, emits laser to target object, then receives the signal reflected by these objects, by comparing the phase difference of received signal and sending signal, the information such as distance, reflectivity of target object can be obtained, through further algorithm processing, the position, height, speed and attitude of target object and other information can also be obtained. Laser radar can effectively avoid the dependence of camera on sunshine intensity and the influence of other sensors such as extreme weather, and the detection of target position can be accurate to centimeter level. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical scheme of the embodiments of the present application, the drawings needed in the embodiments will be briefly introduced as follows, it should be understood that the following drawings only show some embodiments of the present application, therefore should not be regarded as limiting the scope, for ordinary skilled person in the art, on the premise of not paying creative labor, other related drawings can also be obtained according to these drawings.

[0038] Figure 1 The present application provides a kind of based on three-dimensional point cloud vehicle drivable area target identification method and system, by the three-dimensional point cloud of laser radar carried on vehicle emission is cut, to the three-dimensional point cloud ground segmentation after cutting, the drivable area of car is segmented, and non-ground point cloud is reduced sampling, the obstacle in drivable area is detected and identified, the position, speed, attitude, length, width, height, category etc. Information of target is output to decision layer, for the next layer decision algorithm reference.

[0039] Figure 2 The present application provides a kind of based on three-dimensional point cloud vehicle drivable area target identification method and system, by the three-dimensional point cloud of laser radar carried on vehicle emission is cut, to the three-dimensional point cloud ground segmentation after cutting, the drivable area of car is segmented, and non-ground point cloud is reduced sampling, the obstacle in drivable area is detected and identified, the position, speed, attitude, length, width, height, category etc. Information of target is output to decision layer, for the next layer decision algorithm reference.

[0040] Figure 3 The present application provides a kind of based on three-dimensional point cloud vehicle drivable area target identification method and system, by the three-dimensional point cloud of laser radar carried on vehicle emission is cut, to the three-dimensional point cloud ground segmentation after cutting, the drivable area of car is segmented, and non-ground point cloud is reduced sampling, the obstacle in drivable area is detected and identified, the position, speed, attitude, length, width, height, category etc. Information of target is output to decision layer, for the next layer decision algorithm reference.

[0041] Figure 4 The present application provides a kind of based on three-dimensional point cloud vehicle drivable area target identification method and system, by the three-dimensional point cloud of laser radar carried on vehicle emission is cut, to the three-dimensional point cloud ground segmentation after cutting, the drivable area of car is segmented, and non-ground point cloud is reduced sampling, the obstacle in drivable area is detected and identified, the position, speed, attitude, length, width, height, category etc. Information of target is output to decision layer, for the next layer decision algorithm reference. DETAILED DESCRIPTION

[0042] Exemplary embodiments of the present application are described herein with reference to the accompanying drawings, which are cited as illustrative examples. Specific embodiments of the present application are described herein to assist in providing a thorough understanding of embodiments of the application, and are not intended to be limiting in scope. Accordingly, those having ordinary skill in the art will recognize that there are numerous variations and modifications of the embodiments described herein that are within the scope of the present application, and that can be derived from the embodiments described herein without a departing from the scope or spirit of the application. Also, for the purpose of clarity and the brevity, the description below omits the description of well-known functions and structures.

[0043] Obviously, the described embodiments are part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0044] It should be noted that the terminal involved in the embodiments of the present application can include, but is not limited to, a mobile phone, a personal digital assistant (PDA), a wireless handheld device, a tablet computer, a personal computer (PC), an MP3 player, an MP4 player, a wearable device (for example, smart glasses, a smart watch, a smart bracelet, etc.), a smart home device, and the like.

[0045] In addition, the term "and / or" in this paper is only a description of the association relationship between the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " in this paper generally represents that the front and rear associated objects are a "or" relationship.

[0046] The present application will be described in further detail below with reference to the accompanying drawings:

[0047] Referring to Figure 1 , the present application provides a vehicle drivable area target recognition method based on three-dimensional point cloud, comprising the following steps:

[0048] S1: converting the data of the laser radar on the vehicle, converting the angle and distance information in polar coordinates into coordinates in the Cartesian coordinate system.

[0049] Since the data packet of the radar package is only the horizontal rotation angle and the distance parameter, in order to present the effect of the three-dimensional point cloud diagram, as Figure 2 shown, the angle and distance information in polar coordinates in the original data of the laser radar is converted into x, y, z coordinates in the Cartesian coordinate system, and the conversion relationship is

[0050]

[0051] Wherein, r is the measured distance, omega is the vertical angle of the laser, alpha is the horizontal rotation angle of the laser, x, y, z are the coordinates of the polar coordinate projection on the X, Y, Z axes of the Cartesian coordinate system;

[0052] S2: The converted three-dimensional point cloud is cropped, the cropped three-dimensional point cloud is ground segmented, the ground point cloud and the non-ground point cloud are obtained, and the ground point cloud is used as the drivable area.

[0053] The radar installation height is H, and the vehicle body height is h. The vehicle and the obstacles higher than the vehicle have no influence on the vehicle driving, so the points reflected on the vehicle body in the three-dimensional point cloud and all points with z coordinates higher than (h-H) are cropped, and the cropped point cloud is represented by P;

[0054] The laser radar actually emits a cluster of rays, organizes the point cloud in the form of rays, reduces the three-dimensional point cloud (x, y, z) to two-dimensional (x, y), calculates the plane angle theta of each point to the direction in front of the vehicle head, differentiates 360 degrees, and each part has an angle of 0.18 degrees. Each part can be regarded as a ray. The radar distributes multiple lasers inside, emits the radial laser beam as shown in Figure 3 The circle in Figure 3 is the reflection of each cluster of laser beams. After the horizontal angle differentiation of the point cloud, 360 / 0.18=2000 rays can be obtained. The points in these rays are sorted according to the distance;

[0055] Two slope thresholds are set, one is the slope threshold A of the adjacent two points on the same ray, and the other is the slope threshold B of the entire ground. The height threshold is obtained through the two thresholds and the horizontal distance of the current point to the radar. Whether the current point is on the ground is determined by judging whether the z value of the current point is within the height threshold range of the ground. The whole process is shown in Figure 4 The segmented ground point cloud is represented by Pg, and the non-ground point cloud is represented by Png. Meanwhile, the range of Pg is the drivable area of the vehicle;

[0056] S3: The non-ground point cloud is down-sampled, data is collected and labeled, the down-sampled non-ground point cloud is used as input, the labeled label is used as reference, and the detection result is used as output. The neural network model is trained through deep learning.

[0057] The non-ground point cloud is down-sampled, and the non-ground point cloud is segmented by a self-defined size cube. The centroid of the cube is used to represent all the point clouds in the cube, thereby reducing the data amount;

[0058] Collect 3D point cloud data by recording ROS (Robot Operating System) topics with rosbag, and divide the rosbag into single-frame point clouds in bin format according to the sampling frequency;

[0059] Label each frame of the cut-out point cloud with LabelColud, and the label can be customized. Common types include commercial vehicles, passenger cars, motorcycles, electric vehicles, bicycles, pedestrians, and roadblocks.

[0060] Perform deep learning on each frame of the bin-format point cloud and its corresponding label to detect the result as output, train the neural network model, and convert the 3D point cloud into a 2D pseudo-image in the form of Pillar (point column), thereby converting 3D convolution into 2D convolution. The specific implementation steps are as follows:

[0061] Divide the point cloud data into grids according to the X and Y axes (not considering the Z axis), and consider the point cloud data falling into a grid as being in a Pillar. Assuming that there are P non-empty Pillars in each sample, and N point cloud data in each Pillar, the sample can be represented by a (D, P, N) tensor. If the point cloud data in each Pillar exceeds N, it is randomly sampled to N; if the point cloud data in each Pillar is less than N, the missing part is filled with 0; thus, it is easy to convert the point cloud into a (D, P, N) tensor format of stacked Pillars. After obtaining the stacked Pillars, process and feature extract the tensorized point cloud data. Feature extraction can be understood as processing the dimensions of the point cloud. The original point cloud dimension is D = 9, and the processed dimension is C, so a (C, P, N) tensor is obtained. Then, perform a pooling operation according to the Pillar dimension, i.e., obtain a (C, P) dimension feature map. In order to obtain the pseudo-image feature, P is converted to (W, H). Therefore, the final pseudo-image is obtained in the form of (C, H, W). In this way, the 3D to 2D conversion is realized, and then the 2D convolution is performed, and the loss function is sent back to the detection head for online detection.

[0062] S4: Use the trained neural network model for target detection and recognition, extract the target of interest in the drivable area, and publish the target information.

[0063] Use the trained model for target detection and recognition. Only the targets within the vehicle's drivable area will affect the vehicle's driving. Extract the target of interest by comparing whether the target position coincides with the drivable area.

[0064] The position, speed, attitude, length, width, height, category and the like of the target of interest are taken as output and published.

[0065] Embodiment 1

[0066] The embodiment 1 provided by the application is an embodiment of the vehicle drivable area target recognition method based on three-dimensional point cloud provided by the application, and the embodiment of the method comprises the following steps:

[0067] Firstly, since the data packet of the radar package is only the horizontal rotation angle and the distance parameter, in order to present the effect of the three-dimensional point cloud, the angle and distance information in the polar coordinate system need to be converted into the x, y and z coordinates in the Cartesian coordinate system as shown in the figure. Figure 2 Secondly, since the vehicle and the obstacles higher than the vehicle have no influence on the driving of the vehicle, the points reflected on the vehicle body in the converted three-dimensional point cloud and all the points higher than the vehicle height need to be cropped, and then the ground segmentation is performed on the cropped point cloud to segment the ground points and the non-ground points, and the range of the segmented ground points is the drivable area of the vehicle. Next, the non-ground point cloud is down-sampled, the non-ground point cloud is segmented by using the self-defined size of the cube, and the centroid of the cube is used to represent all the point clouds of the cube. Then, the data collection and labeling are performed, the labeling label can be self-defined, and the commonly used labeling label types include commercial vehicles, passenger vehicles, motorcycles, electric vehicles, bicycles, pedestrians, roadblocks and the like. With the down-sampled non-ground point cloud as the input, the labeling label as the reference, and the detection result as the output, the deep learning is performed, the neural network model is trained, the trained model is used for target detection and recognition, the interested target is extracted through the segmented drivable area, finally the position, length, width, height, category and the like of the target are taken as the output and published, the function of target recognition is realized, and the output information can be referred to by the next layer of decision algorithm.

[0068] Embodiment 2

[0069] The embodiment 2 provided by the application is an embodiment of the vehicle drivable area target recognition system based on three-dimensional point cloud provided by the application, and the embodiment of the system comprises a conversion module, a segmentation module, a down-sampling module and an extraction module.

[0070] The conversion module is used for converting the data of the laser radar on the vehicle, converting the angle and distance information in the polar coordinate system into the coordinates in the Cartesian coordinate system;

[0071] The segmentation module is used for cropping the converted three-dimensional point cloud, performing ground segmentation on the cropped three-dimensional point cloud to obtain the ground point cloud and the non-ground point cloud, and taking the ground point cloud as the drivable area;

[0072] The downsampling module is used for downsampling the non-ground point cloud, collecting data and labeling tags, taking the downsampled non-ground point cloud as input, taking the labeled tags as reference, taking the detection result as output, and training the neural network model through deep learning.

[0073] The extraction module is used for target detection and identification by using the trained neural network model, extracting the interested target of the drivable area, and publishing target information.

[0074] It can be understood that the vehicle drivable area target identification system based on the three-dimensional point cloud provided by the present application corresponds to the vehicle drivable area target identification method based on the three-dimensional point cloud provided by the above-mentioned embodiments, and the related technical features of the vehicle drivable area target identification system based on the three-dimensional point cloud can refer to the related technical features of the vehicle drivable area target identification method based on the three-dimensional point cloud, which will not be repeated here.

[0075] The present embodiment provides a terminal device, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to realize the following steps: converting the data of the laser radar, converting the angle and distance information in polar coordinates into coordinates in the Cartesian coordinate system; cropping the converted three-dimensional point cloud, ground segmenting the cropped three-dimensional point cloud to obtain ground point cloud and non-ground point cloud; downsampling the non-ground point cloud, collecting data and labeling, taking the downsampled non-ground point cloud as input, taking the labeled tags as reference, performing deep learning, and outputting a neural network model; using the trained neural network model for target detection and identification, and extracting the interested target of the segmented drivable area, and publishing target information.

[0076] The present embodiment provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the following steps: converting the data of the laser radar, converting the angle and distance information in polar coordinates into coordinates in the Cartesian coordinate system; cropping the converted three-dimensional point cloud, ground segmenting the cropped three-dimensional point cloud to obtain ground point cloud and non-ground point cloud; downsampling the non-ground point cloud, collecting data and labeling, taking the downsampled non-ground point cloud as input, taking the labeled tags as reference, performing deep learning, and outputting a neural network model; using the trained neural network model for target detection and identification, and extracting the interested target of the segmented drivable area, and publishing target information.

[0077] The embodiment of the present application provides a vehicle drivable area target identification method and system based on three-dimensional point cloud, terminal equipment and storage medium, through conversion of data of a laser radar, angle and distance information in polar coordinates are converted into coordinates in a Cartesian coordinate system. Secondly, the converted three-dimensional point cloud is cropped, and the point cloud in the area that is too high and reflected on the vehicle body can be ignored, then the ground of the cropped three-dimensional point cloud is segmented to obtain ground point cloud and non-ground point cloud, and the range of the segmented ground point cloud is the drivable area. Next, the non-ground point cloud is down-sampled, the non-ground point cloud is segmented by using a self-defined size cube, and the centroid of the cube is used to represent all point clouds of the cube. Then data collection and labeling are performed, and the labeling label can be self-defined, and common types include commercial vehicles, passenger vehicles, motorcycles, electric vehicles, bicycles, pedestrians, roadblocks and the like. Taking the down-sampled non-ground point cloud as input and the labeling label as reference, deep learning is performed, a neural network model is output, target detection and identification are performed by using the trained model, interested targets are extracted through the segmented drivable area, and finally the position, length, width, height and category of the target are taken as output and published for reference of a next layer decision algorithm.

[0078] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, system or computer program product. Therefore, the present application can adopt a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.

[0079] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system) and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be realized by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one flow or multiple flows and / or blocks Figure 1 The device for implementing the functions specified in one block or multiple blocks.

[0080] These computer program instructions can also be stored in a computer readable storage medium capable of guiding a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable storage medium produce a product including instruction devices, which implement the functions specified in the flowcharts and / or block diagrams.Figure 1 one or more processes and / or functions described in the one or more blocks. Figure 1 one or more blocks.

[0081] These computer program instructions can also be loaded into computer or other programmable data processing devices, so that a series of operation steps are performed on the computer or other programmable data processing devices to generate computer-implemented processes, so that the instructions executed on the computer or other programmable data processing devices provide steps for implementing the processes described in the one or more blocks. Figure 1 one or more processes and / or functions described in the one or more blocks. Figure 1 one or more blocks.

[0082] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, but not to limit it, although the present application has been described in detail with reference to the above examples, those skilled in the art should understand: the specific embodiments of the present application can still be modified or replaced by the equivalent, without departing from the spirit and scope of the present application, any modification or equivalent replacement, which should be covered in the protection scope of the claims of the present application.

Claims

1. A method for target recognition of a drivable area of a vehicle based on a three-dimensional point cloud, characterized in that The method comprises the following steps: Converting data of a laser radar on a vehicle, converting angle and distance information in polar coordinates into coordinates in a Cartesian coordinate system; Carrying out ground segmentation on the cropped three-dimensional point cloud to obtain a ground point cloud and a non-ground point cloud, and taking the ground point cloud as a drivable area; Carrying out data collection and label annotation on the non-ground point cloud, taking the non-ground point cloud after down-sampling as input, taking the label annotation as reference, and taking a detection result as output, and training a neural network model through deep learning; Using the trained neural network model to carry out target detection and recognition, extracting a target of interest in the drivable area, and publishing target information; The neural network model converts three-dimensional point cloud into a two-dimensional pseudo image in the form of Pillar, thereby converting 3D convolution into 2D convolution, and the specific implementation steps are as follows: Dividing point cloud data into grids according to the X and Y axes of the point cloud data, and regarding point cloud data falling into a grid as being in a Pillar; Supposing that there are P non-empty Pillars in each sample, and there are N point cloud data in each Pillar, the sample represents a (D, P, N) tensor, if the point cloud data in each Pillar exceeds N, the point cloud data is randomly sampled to N, if the point cloud data in each Pillar is less than N, the part less than N is filled with 0, and the point cloud is converted into a (D, P, N) tensor format of stacked Pillars; After obtaining the stacked Pillars, processing and feature extraction are carried out on the tensorized point cloud data, a (C, P) dimensional feature map is obtained through a pooling operation according to the dimension of the Pillar, P is converted into (W, H), and finally a pseudo image in the form of (C, H, W) is obtained. 2.The three-dimensional point cloud based vehicle drivable region target identification method of claim 1, wherein, The conversion relationship of the angle and distance information in polar coordinates into coordinates in the Cartesian coordinate system is wherein, is the measured distance, is the vertical angle of the laser, is the horizontal rotation angle of the laser, is the polar projection onto the Cartesian coordinate system axis. 3.The three-dimensional point cloud based vehicle drivable region target identification method of claim 1, wherein, The cropping comprises cropping points reflected on the vehicle body and all points higher than the vehicle height in the converted three-dimensional point cloud. 4.The three-dimensional point cloud based vehicle drivable region target identification method of claim 1, wherein, The ground segmentation process is as follows: The laser radar emits a cluster of rays, reduces the three-dimensional point cloud (x, y, z) to two dimensions (x, y), calculates the angle θ of each point to the plane in the direction of the front of the vehicle head, differentiates 360°, and each part has an angle of 0.18°, which can be regarded as a ray, and the points in these rays are sorted according to the distance; Two slope thresholds are set, one is the slope threshold A of adjacent two points on the same ray, and the other is the slope threshold B of the entire ground, the height threshold is obtained through the two thresholds and the horizontal distance of the current point to the radar, and whether the current point is on the ground is determined by judging whether the z value of the current point is within the height threshold range. 5.The three-dimensional point cloud based vehicle drivable region target identification method of claim 1, wherein, The down-sampling is to divide the non-ground point cloud into cubes of a self-defined size, and use the center of the cube to represent all point clouds in the cube. 6.The three-dimensional point cloud based vehicle drivable region target identification method of claim 1, wherein, The data collection and label annotation process is as follows: The three-dimensional point cloud data is collected by recording the ROS topics through the rosbag, and the rosbag is divided into single-frame point clouds in the bin format according to the sampling frequency; Each frame of the cut-out point cloud is labeled by LabelColud, and the label includes commercial vehicles, passenger vehicles, motorcycles, electric vehicles, bicycles, pedestrians and roadblocks.

7. A vehicle drivable area object recognition system based on three-dimensional point clouds, characterized by Steps for implementing the method of claim 1, comprising: A conversion module is configured to convert data of a laser radar on a vehicle, and convert angle and distance information in polar coordinates into coordinates in a Cartesian coordinate system; A segmentation module is configured to crop the converted three-dimensional point cloud, ground segment the cropped three-dimensional point cloud, and obtain a ground point cloud and a non-ground point cloud, and take the ground point cloud as a drivable area; A down-sampling module is configured to down-sample the non-ground point cloud, collect data and label, take the down-sampled non-ground point cloud as an input, take the label as a reference, take a detection result as an output, and train a neural network model through deep learning; An extraction module is configured to detect and identify a target by using the trained neural network model, extract a target of interest in the drivable area, and publish target information.

8. A terminal device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The computer program is executed by the processor to implement the steps of the method of any one of claims 1-6.

Images