Obstacle heading angle acquisition method, device, electronic device, vehicle and computer-readable storage medium

Abstract

The present application provides a method, device, electronic device, vehicle, and computer-readable storage medium for obtaining an obstacle heading angle. The method includes: detecting a bird's-eye view of a final obstacle point cloud set using a convolutional neural network model to determine an initial 2D detection frame of the bird's-eye view; converting the initial 2D detection frame into an initial 3D detection frame in a radar coordinate system, and performing cluster analysis on the point cloud set within the initial 3D detection frame to obtain multiple point cloud clusters; determining a target point cloud cluster from the multiple point cloud clusters based on the number of three-dimensional points in each point cloud cluster, and obtaining a target 3D detection frame corresponding to the target point cloud cluster; obtaining the speed of the target 3D detection frame using a multi-target tracking algorithm, and determining the obstacle heading angle based on the speed of the target 3D detection frame. In this way, end-to-end reasoning can be achieved, and the obstacle yaw angle can be output in one go, thereby improving the accuracy of the obstacle heading angle.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving technology, and in particular to a method, device, electronic device, vehicle, and computer-readable storage medium for obtaining an obstacle heading angle. Background Art

[0002] Currently, there are two main approaches to implementing autonomous driving technology: one uses the vehicle's own computing platform and sensors to perceive the surrounding environment, and the other uses vehicle-road collaboration. This approach allows for beyond-visual-range perception. Roadside units (ROUs) use sensors to accurately transmit road information that the vehicle cannot perceive, reducing blind spots. This approach requires detecting and identifying obstacles at intersections and predicting their characteristic information, such as their heading angle.

[0003] Existing obstacle heading angle prediction technologies mainly include the following two parts: (1) point cloud obstacle detection based on Euclidean clustering; (2) obstacle heading angle prediction based on principal component analysis. The use of principal component analysis to predict obstacle yaw angles is easily affected by the distribution of point clouds. For example, a vehicle with a long radar scanning distance cannot completely scan the entire outline of the point cloud, resulting in inaccurate obstacle heading angles obtained using principal component analysis. Summary of the Invention

[0004] In order to solve the above technical problems, embodiments of the present application provide a method, device, electronic device, vehicle and computer-readable storage medium for obtaining an obstacle heading angle.

[0005] In a first aspect, an embodiment of the present application provides a method for obtaining an obstacle heading angle, the method comprising:

[0006] Detecting the final bird's-eye view of the obstacle point cloud using a convolutional neural network model to determine an initial 2D detection frame of the bird's-eye view;

[0007] Converting the initial 2D detection frame into an initial 3D detection frame in a radar coordinate system, and performing cluster analysis on the point cloud set within the initial 3D detection frame to obtain multiple point cloud clusters;

[0008] Determine a target point cloud cluster from the multiple point cloud clusters according to the number of three-dimensional points in each point cloud cluster, and obtain a target 3D detection frame corresponding to the target point cloud cluster;

[0009] The speed of the target 3D detection frame is obtained through a multi-target tracking algorithm, and the heading angle of the obstacle is determined according to the speed of the target 3D detection frame.

[0010] In a second aspect, an embodiment of the present application provides a device for obtaining an obstacle heading angle, the device comprising:

[0011] A detection module is used to detect the final bird's-eye view of the obstacle point cloud set through a convolutional neural network model and determine an initial 2D detection frame of the bird's-eye view;

[0012] a conversion module, configured to convert the initial 2D detection frame into an initial 3D detection frame in a radar coordinate system, and perform cluster analysis on the point cloud set within the initial 3D detection frame to obtain a plurality of point cloud clusters;

[0013] an acquisition module, configured to determine a target point cloud cluster from the plurality of point cloud clusters according to the number of three-dimensional points in each point cloud cluster, and to obtain a target 3D detection frame corresponding to the target point cloud cluster;

[0014] The determination module is used to obtain the speed of the target 3D detection frame through a multi-target tracking algorithm, and determine the heading angle of the obstacle according to the speed of the target 3D detection frame.

[0015] In a third aspect, an embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and when the computer program is run by the processor, the method for obtaining the obstacle heading angle provided in the first aspect is executed.

[0016] In a fourth aspect, an embodiment of the present application provides a vehicle, comprising a vehicle body and the electronic device provided in the third aspect.

[0017] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium storing a computer program, which, when running on a processor, executes the method for obtaining the obstacle heading angle provided in the first aspect.

[0018] The obstacle heading angle acquisition method, device, electronic device, vehicle, and computer-readable storage medium provided by the present application detect the bird's-eye view of the final obstacle point cloud set through a convolutional neural network model to determine the initial 2D detection frame of the bird's-eye view; convert the initial 2D detection frame into an initial 3D detection frame in the radar coordinate system, and perform cluster analysis on the point cloud set within the initial 3D detection frame to obtain multiple point cloud clusters; determine the target point cloud cluster from the multiple point cloud clusters based on the number of three-dimensional points in each point cloud cluster, and obtain the target 3D detection frame corresponding to the target point cloud cluster; obtain the speed of the target 3D detection frame through a multi-target tracking algorithm, and determine the obstacle heading angle based on the speed of the target 3D detection frame. In this way, end-to-end reasoning can be achieved, and the obstacle yaw angle can be output at one time. This can effectively solve the problem that the object heading angle calculation using principal component analysis is easily affected by the object point cloud distribution and point cloud completeness, thereby improving the accuracy of the obstacle heading angle. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solution of this application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of this application and should not be regarded as limiting the scope of protection of this application. In each of the drawings, similar components are numbered similarly.

[0020] Figure 1 A schematic diagram showing a flow chart of a method for obtaining an obstacle heading angle provided in an embodiment of the present application is shown;

[0021] Figure 2 A schematic diagram of an AABB-type 2D detection frame provided in an embodiment of the present application is shown;

[0022] Figure 3 A schematic structural diagram of an obstacle heading angle acquisition device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0024] The components of the embodiments of the present application generally described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but rather merely represents selected embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0025] Hereinafter, the terms "including", "having" and their cognates, which may be used in various embodiments of the present application, are intended only to indicate specific features, numbers, steps, operations, elements, components or combinations of the foregoing items, and should not be understood as first excluding the existence of one or more other features, numbers, steps, operations, elements, components or combinations of the foregoing items or the possibility of adding one or more features, numbers, steps, operations, elements, components or combinations of the foregoing items.

[0026] Furthermore, the terms “first,” “second,” “third,” etc., are merely used for distinguishing descriptions and are not to be understood as indicating or implying relative importance.

[0027] Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art to which the various embodiments of the present application belong. The terms (such as those defined in generally used dictionaries) will be interpreted as having the same meaning as in the context of the relevant technical field and will not be interpreted as having an idealized meaning or an overly formal meaning unless clearly defined in the various embodiments of the present application.

[0028] Example 1

[0029] An embodiment of the present disclosure provides a method for obtaining an obstacle heading angle.

[0030] For details, see Figure 1 , the method for obtaining the obstacle heading angle includes:

[0031] Step S101: Detect the final bird's-eye view of the obstacle point cloud set through a convolutional neural network model to determine the initial 2D detection frame of the bird's-eye view.

[0032] The method for obtaining the obstacle heading angle in this embodiment can be applied to vehicles equipped with laser radars, or to electronic devices. The acquisition process of the convolutional neural network model is as follows: a deep learning method is used to annotate the bird's-eye view training set, and the annotated bird's-eye view training set is input into the convolutional neural network model to be trained. The convolutional neural network model detects the initial 2D detection frame in the bird's-eye view of the final obstacle point cloud set, and can also determine the category of each initial 2D detection frame, for example, each initial 2D detection frame is a motor vehicle category, a bicycle category, etc. The 2D detection frame can also be called a 2D bounding box.

[0033] In this embodiment, the convolutional neural network model detects a rough 2D bounding box in the bird's-eye view, and the detected 2D detection box is of AABB type. If the distance between the objects to be detected is close, the 2D detection boxes will cross. Figure 2 ,An AABB-type 2D detection frame means that two opposite sides in the detection frame are parallel to the x-axis of the two-dimensional plane, and the other two opposite sides are parallel to the y-axis of the two-dimensional plane.

[0034] In this embodiment, obtaining the final obstacle point cloud includes:

[0035] Acquire original point cloud data, and delete the ground point cloud set from the original point cloud data to obtain an initial obstacle point cloud set;

[0036] The initial obstacle point cloud set is downsampled according to a preset three-dimensional voxel grid to obtain the final obstacle point cloud set.

[0037] In this embodiment, a laser radar can be used to obtain original point cloud data of the surrounding environment. Since the laser radar has an effective scanning distance, the original point cloud data within the effective scanning distance of the surrounding environment where the laser radar is located can be obtained.

[0038] In this embodiment, the number of 3D points in the raw point cloud data obtained by LiDAR scanning is relatively large, typically measured in tens of thousands. Processing all 3D points would require significant computing resources. Downsampling the initial obstacle point cloud set according to a preset 3D voxel grid reduces the number of 3D points while preserving the shape characteristics of the point cloud data and improving the algorithm's execution rate. By creating corresponding 3D voxel grids based on the input raw point cloud data according to the preset 3D voxel grid, the position coordinates of the center of gravity of all 3D points in each 3D voxel grid are calculated, and the position coordinates of the center of gravity are used to represent all 3D points in the 3D voxel grid. Thus, all points in the 3D voxel grid are represented by a single center of gravity point.

[0039] Specifically, lidar can be used to perform low-level obstacle detection. Considering factors such as hardware performance, development cycle, and data costs, point cloud processing algorithms can be used to segment, fit, and track obstacles. In one embodiment, a clustering algorithm can be used for obstacle segmentation. Specifically, attributes can be measured based on the Euclidean distance between points. Therefore, 3D points on the ground will interfere with the 3D point set of objects located on the ground.

[0040] In this embodiment, the acquisition of the ground point cloud set includes:

[0041] Randomly selecting a preset number of three-dimensional point sets from the original point cloud data, and determining a corresponding plane equation based on each three-dimensional point set;

[0042] Counting the number of interior points belonging to each plane equation in the original point cloud data;

[0043] The plane equation with the largest number of interior points is taken as the ground equation, and the set of all interior points of the ground equation is taken as the ground point cloud set.

[0044] In this embodiment, a random sample consensus (RANSAC) algorithm can be used to determine the ground equation through plane fitting. Specifically, the following steps are included: a preset number of 3D points are randomly selected to determine a plane equation. For example, three 3D points are selected, and the values ​​of the parameters of Formula 1 are determined. Formula 1: Ax+By+Cz+D=0. The parameter values ​​of Formula 1 are calculated by setting parameter D to 0, substituting the coordinate values ​​of the three 3D points in the first, second, and third directions as the values ​​of x, y, and z, respectively, into Formula 1 to obtain three equations. Parameters A, B, and C are calculated based on the three equations, and the calculated parameters A, B, and C are substituted into Formula 1 to obtain a plane equation. The above steps of obtaining a plane equation are repeated a preset number of times to obtain a preset number of plane equations. Substitute all points into each plane equation in turn, and determine whether each three-dimensional point is an interior point of each plane equation based on the preset distance threshold. If the coordinate values ​​of the first, second, and third directions of the three-dimensional point are substituted into each plane equation as the values ​​of x, y, and z respectively, the calculated parameter D is determined whether the calculated parameter D is less than or equal to the preset threshold. If the calculated parameter D is less than or equal to the preset threshold, then the three-dimensional point is an interior point of the plane equation. If the calculated parameter D is greater than the preset threshold, then the three-dimensional point is an exterior point of the plane equation. Traverse all three-dimensional points to determine whether each three-dimensional point is an interior point or exterior point of the plane equation, and count the number of interior points of the plane equation; determine the plane equation with the largest number of interior points as the ground equation, determine the set of all interior points of the ground equation as the ground point cloud set, and the exterior points of the ground equation are the obstacle point cloud set.

[0045] In this embodiment, obtaining the final bird's-eye view of the obstacle point cloud includes:

[0046] The final obstacle point cloud set is projected onto a two-dimensional plane according to a preset projection density to obtain the bird's-eye view.

[0047] In this embodiment, the unit of the preset projection density can be m / pixel. The preset projection density means that one pixel on the image represents a preset distance in the real world. The preset projection density can affect the feature sparsity of the bird's-eye view image and the subsequent convolutional neural network recognition effect. After obtaining the bird's-eye view image, the convolutional neural network model can be used to detect and recognize the bird's-eye view image.

[0048] Step S102 : converting the initial 2D detection frame into an initial 3D detection frame in a radar coordinate system, and performing cluster analysis on the point cloud set within the initial 3D detection frame to obtain a plurality of point cloud clusters.

[0049] In this embodiment, the initial 2D detection frame is a rough detection frame detected in the bird's-eye view. The initial 2D detection frame can also be called an initial 2D bounding box. The initial 2D detection frame is a 2D detection frame in the bird's-eye view pixel coordinate system. The conversion relationship between the bird's-eye view pixel coordinate system and the lidar coordinate system can be obtained. Based on this conversion relationship, the initial 2D detection frame is converted to lidar coordinates to obtain the corresponding 2D detection frame in the radar coordinate system. The initial 3D detection frame in the radar coordinate system is determined based on the corresponding 2D detection frame in the radar coordinate system and the height of the preset obstacle.

[0050] Specifically, the converting of the initial 2D detection frame into an initial 3D detection frame in the radar coordinate system in step S102 includes:

[0051] Converting the initial 2D detection frame into a revised 2D detection frame in a radar coordinate system;

[0052] An initial 3D detection frame in the radar coordinate system is determined based on the modified 2D detection frame and the preset obstacle height.

[0053] In this embodiment, the revised 2D detection frame in the radar coordinate system can be represented by the position coordinates in the first direction and the second direction in the radar coordinate system. For example, the x- and y-direction coordinates of the obstacle in the lidar coordinate system. The preset obstacle height can be a user-defined obstacle height, which can be determined based on the data distribution in the bird's-eye view training set. Based on the obstacle height, the radar coordinate system estimates the position coordinates of the obstacle in the first direction. Specifically, the position coordinates in the third direction can be the z-direction coordinates of the obstacle in the lidar coordinate system. The x-, y-, and z-direction coordinates of the obstacle in the lidar coordinate system are used to determine the initial 3D detection frame of the obstacle in the lidar coordinate system.

[0054] In this embodiment, obtaining a point cloud set within an initial 3D detection frame includes the following steps:

[0055] The final obstacle point cloud set is segmented according to the initial 3D detection frame, and three-dimensional points outside the initial 3D detection frame are deleted to obtain a point cloud set within the initial 3D detection frame.

[0056] In this embodiment, the initial 3D detection frame is used as a conditional filter to segment all 3D points within the initial 3D detection frame from the standard obstacle point cloud. All segmented 3D points are used as a roughly segmented obstacle point cloud set. The roughly segmented obstacle point cloud is used as the input of the subsequent algorithm to obtain a more accurate obstacle point cloud from the roughly segmented obstacle point cloud set.

[0057] In this embodiment, in step S102, cluster analysis is performed on the point cloud set within the initial 3D detection frame to obtain multiple point cloud clusters, including:

[0058] Perform Euclidean clustering analysis on all point clouds within the initial 3D detection frame to obtain multiple point cloud clusters.

[0059] In this embodiment, the Euclidean clustering analysis includes: examining all m three-dimensional points in the initial 3D detection frame, defining affinity clusters with preset properties between the three-dimensional points in the m-dimensional space, assuming that the m three-dimensional points constitute n classes, and then merging the two classes with the smallest distance into one class, and recalculating the distance between classes, iterating until the distance between any two classes is greater than a specified threshold, or the number of classes is less than a specified number, and completing the segmentation to obtain multiple point cloud clusters.

[0060] Step S103 : determining a target point cloud cluster from the multiple point cloud clusters according to the number of three-dimensional points in each point cloud cluster, and obtaining a target 3D detection frame corresponding to the target point cloud cluster.

[0061] In this embodiment, statistical clustering is performed to determine the number of 3D points in each point cloud cluster. The clusters are then sorted by the number of 3D points. The cluster with the largest number of 3D points is selected as the target 3D detection box. The 3D detection box of this target point cloud cluster is the detection box that best fits the obstacle. The 3D detection box is also called a 3D bounding box.

[0062] Step S104 : obtaining the speed of the target 3D detection frame through a multi-target tracking algorithm, and determining the heading angle of the obstacle according to the speed of the target 3D detection frame.

[0063] In one embodiment, the size, location, and category of the obstacle can also be determined based on the target 3D detection frame. The size information of the obstacle includes information such as the length, width, and height of the obstacle. The location information of the obstacle may include three-dimensional coordinate information, and the category of the obstacle may include bicycles, buses, cars, etc.

[0064] In this embodiment, the target 3D detection frame obtained by clustering is input into the multi-target tracking algorithm. The target tracking algorithm can be used to calculate the speed of the obstacle on the two-dimensional plane. Specifically, the speed of the obstacle on the XOY plane can be calculated, and the heading angle of the obstacle can be calculated based on the speed components of the obstacle on the x-axis and y-axis. In this way, the yaw angle is calculated using the speed information obtained by the target tracking algorithm, which is not affected by the integrity, distribution, and shape of the object point cloud. The calculated yaw angle is more stable, which can effectively solve the problem that the heading angle of the object calculated using principal component analysis is easily affected by the distribution of the object point cloud and the integrity of the point cloud, thereby improving the accuracy of the obstacle's heading angle. In this embodiment, end-to-end reasoning can be achieved, and the size information, position information, and yaw angle information of the obstacle can be output at one time, thereby improving the accuracy of obtaining various feature information of the obstacle.

[0065] The method for obtaining the obstacle heading angle provided in this embodiment detects the bird's-eye view of the final obstacle point cloud set through a convolutional neural network model to determine the initial 2D detection frame of the bird's-eye view; converts the initial 2D detection frame into an initial 3D detection frame in the radar coordinate system, and performs cluster analysis on the point cloud set within the initial 3D detection frame to obtain multiple point cloud clusters; determines the target point cloud cluster from the multiple point cloud clusters based on the number of three-dimensional points in each point cloud cluster, and obtains the target 3D detection frame corresponding to the target point cloud cluster; obtains the speed of the target 3D detection frame through a multi-target tracking algorithm, and determines the obstacle heading angle based on the speed of the target 3D detection frame. In this way, end-to-end reasoning can be achieved, and the size information, position information, and yaw angle information of the obstacle can be output at one time. This can effectively solve the problem that the calculation of the object heading angle using principal component analysis is easily affected by the distribution of the object point cloud and the completeness of the point cloud, thereby improving the accuracy of the obstacle heading angle.

[0066] Example 2

[0067] In addition, an embodiment of the present disclosure provides a device for acquiring an obstacle heading angle.

[0068] Specifically, such as Figure 3 As shown, the obstacle heading angle acquisition device 300 includes:

[0069] A detection module 301 is configured to detect the final obstacle point cloud in the bird's-eye view using a convolutional neural network model to determine an initial 2D detection frame of the bird's-eye view.

[0070] a conversion module 302 for converting the initial 2D detection frame into an initial 3D detection frame in a radar coordinate system, and performing cluster analysis on the point cloud set within the initial 3D detection frame to obtain a plurality of point cloud clusters;

[0071] A first determining module 303 is configured to determine a target point cloud cluster from the plurality of point cloud clusters according to the number of three-dimensional points in each point cloud cluster, and obtain a target 3D detection frame corresponding to the target point cloud cluster;

[0072] The second determining module 304 is configured to obtain the velocity of the target 3D detection frame by using a multi-target tracking algorithm, and determine the heading angle of the obstacle according to the velocity of the target 3D detection frame.

[0073] In this embodiment, the conversion module 302 is further configured to convert the initial 2D detection frame into a modified 2D detection frame in the radar coordinate system;

[0074] An initial 3D detection frame in the radar coordinate system is determined based on the modified 2D detection frame and the preset obstacle height.

[0075] In this embodiment, the obstacle heading angle acquisition device 300 further includes:

[0076] The first acquisition module is configured to segment the final obstacle point cloud set according to the initial 3D detection frame, delete three-dimensional points outside the initial 3D detection frame, and obtain a point cloud set within the initial 3D detection frame.

[0077] In this embodiment, the obstacle heading angle acquisition device 300 further includes:

[0078] a second acquisition module, configured to acquire original point cloud data and delete a ground point cloud set from the original point cloud data to obtain an initial obstacle point cloud set;

[0079] The initial obstacle point cloud set is downsampled according to a preset three-dimensional voxel grid to obtain the final obstacle point cloud set.

[0080] In this embodiment, the obstacle heading angle acquisition device 300 further includes:

[0081] a third acquisition module, configured to randomly select a preset number of three-dimensional point sets from the original point cloud data, and determine a corresponding plane equation based on each three-dimensional point set;

[0082] Counting the number of interior points belonging to each plane equation in the original point cloud data;

[0083] The plane equation with the largest number of interior points is taken as the ground equation, and the set of all interior points of the ground equation is taken as the ground point cloud set.

[0084] In this embodiment, the obstacle heading angle acquisition device 300 further includes:

[0085] The fourth acquisition module is used to project the final obstacle point cloud set onto a two-dimensional plane according to a preset projection density to obtain the bird's-eye view.

[0086] The obstacle heading angle acquisition device provided in this embodiment detects the bird's-eye view of the final obstacle point cloud set through a convolutional neural network model to determine the initial 2D detection frame of the bird's-eye view; converts the initial 2D detection frame into an initial 3D detection frame in the radar coordinate system, and performs cluster analysis on the point cloud set within the initial 3D detection frame to obtain multiple point cloud clusters; determines the target point cloud cluster from the multiple point cloud clusters based on the number of three-dimensional points in each point cloud cluster, and obtains the target 3D detection frame corresponding to the target point cloud cluster; obtains the speed of the target 3D detection frame through a multi-target tracking algorithm, and determines the obstacle heading angle based on the speed of the target 3D detection frame. In this way, end-to-end reasoning can be achieved, and the size information, position information and yaw angle information of the obstacle can be output at one time. This can effectively solve the problem that when using principal component analysis to calculate the object heading angle, it is easily affected by the distribution of the object point cloud and the completeness of the point cloud, thereby improving the accuracy of the obstacle's feature information.

[0087] Example 3

[0088] In addition, an embodiment of the present disclosure provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program, the method for obtaining the obstacle heading angle provided in Example 1 is executed.

[0089] The electronic device provided in this embodiment can execute the steps of the method for obtaining the obstacle heading angle provided in Example 1, and to avoid repetition, they are not described here.

[0090] Example 4

[0091] In addition, an embodiment of the present disclosure provides a vehicle, including a vehicle body and the electronic device provided in Example 3.

[0092] The vehicle provided in this embodiment can execute the steps of the method for obtaining the obstacle heading angle provided in Example 1, which will not be described again here to avoid repetition.

[0093] Example 5

[0094] In addition, an embodiment of the present disclosure provides a computer-readable storage medium storing a computer program. When the computer program is run on a processor, the method for obtaining the obstacle heading angle provided in embodiment 1 is executed.

[0095] In this embodiment, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0096] The computer-readable storage medium provided in this embodiment can execute the steps of the method for obtaining the obstacle heading angle provided in Example 1, and to avoid repetition, they are not described here.

[0097] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or terminal comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or terminal comprising the element.

[0098] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0099] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

Claims

1. A method for obtaining an obstacle heading angle, characterized in that: The method comprises: Detecting the final bird's-eye view of the obstacle point cloud using a convolutional neural network model to determine an initial 2D detection frame of the bird's-eye view; Converting the initial 2D detection frame into an initial 3D detection frame in a radar coordinate system, and performing cluster analysis on the point cloud set within the initial 3D detection frame to obtain multiple point cloud clusters; Determine a target point cloud cluster from the multiple point cloud clusters according to the number of three-dimensional points in each point cloud cluster, and obtain a target 3D detection frame corresponding to the target point cloud cluster; Obtaining the speed of the target 3D detection frame through a multi-target tracking algorithm, and determining the heading angle of the obstacle based on the speed of the target 3D detection frame; The converting the initial 2D detection frame into an initial 3D detection frame in the radar coordinate system includes: Converting the initial 2D detection frame into a modified 2D detection frame in the radar coordinate system; the modified 2D detection frame is represented by the coordinate values ​​of the obstacle in the x direction and the coordinate values ​​of the obstacle in the y direction in the radar coordinate system; An initial 3D detection frame in the radar coordinate system is determined based on the modified 2D detection frame and the preset obstacle height.

2. The method according to claim 1, characterized in that Acquiring the point cloud set within the initial 3D detection frame includes: The final obstacle point cloud set is segmented according to the initial 3D detection frame, and three-dimensional points outside the initial 3D detection frame are deleted to obtain a point cloud set within the initial 3D detection frame.

3. The method according to claim 1, characterized in that The acquisition of the final obstacle point cloud set includes: Acquire original point cloud data, and delete the ground point cloud set from the original point cloud data to obtain an initial obstacle point cloud set; The initial obstacle point cloud set is downsampled according to a preset three-dimensional voxel grid to obtain the final obstacle point cloud set.

4. The method according to claim 3, characterized in that The acquisition of the ground point cloud set includes: Randomly selecting a preset number of three-dimensional point sets from the original point cloud data, and determining a corresponding plane equation based on each three-dimensional point set; Counting the number of interior points belonging to each plane equation in the original point cloud data; The plane equation with the largest number of interior points is taken as the ground equation, and the set of all interior points of the ground equation is taken as the ground point cloud set.

5. The method according to claim 1, wherein The acquisition of the final bird's-eye view of the obstacle point cloud includes: The final obstacle point cloud set is projected onto a two-dimensional plane according to a preset projection density to obtain the bird's-eye view.

6. A device for obtaining an obstacle heading angle, characterized in that: The device comprises: A detection module is used to detect the final bird's-eye view of the obstacle point cloud set through a convolutional neural network model and determine an initial 2D detection frame of the bird's-eye view; a conversion module, configured to convert the initial 2D detection frame into an initial 3D detection frame in a radar coordinate system, and perform cluster analysis on the point cloud set within the initial 3D detection frame to obtain a plurality of point cloud clusters; A first determination module is configured to determine a target point cloud cluster from the plurality of point cloud clusters according to the number of three-dimensional points in each point cloud cluster, and obtain a target 3D detection frame corresponding to the target point cloud cluster; A second determination module is configured to obtain a velocity of the target 3D detection frame by using a multi-target tracking algorithm, and determine a heading angle of the obstacle according to the velocity of the target 3D detection frame; The conversion module is further configured to convert the initial 2D detection frame into a modified 2D detection frame in a radar coordinate system; the modified 2D detection frame is represented by coordinate values ​​of the obstacle in the x-direction and the y-direction in the radar coordinate system; An initial 3D detection frame in the radar coordinate system is determined based on the modified 2D detection frame and the preset obstacle height.

7. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program, the method for obtaining the obstacle heading angle according to any one of claims 1 to 5 is executed.

8. A vehicle comprising a vehicle body, characterized in that: Also included is an electronic device as claimed in claim 7.

9. A computer-readable storage medium, characterized in that The device stores a computer program, which, when running on a processor, executes the method for obtaining the obstacle heading angle according to any one of claims 1 to 5.

Images