A method and apparatus for detecting a target object

Abstract

The application discloses a kind of detection methods and devices of target object, it is related to computer technical field.The specific embodiment of the method includes: to each point cloud point of original point cloud data, first coordinate transformation is carried out to point cloud point, and first feature extraction is carried out to obtain first local feature, and first local feature includes spatial position information;Second coordinate transformation is carried out to point cloud point, and second feature extraction is carried out to obtain second local feature, and second local feature includes height position information;First local feature and second local feature are fused, and the target local feature of point cloud point is obtained;Global feature of point cloud point is obtained by using neural network to carry out multilayer perception learning to target local feature;Global feature is input into target detection model, and the detection result of target object is obtained.Through twice coordinate transformation to point cloud point, the accuracy of target object detection is improved, and target object is better identified.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to a target object detection method and device. BACKGROUND

[0002] 3D target object detection is a very important technology in the field of automatic driving. By detecting and identifying obstacles that hinder driving, a reasonable avoidance plan is made for different obstacles according to the detection result to ensure the safety of vehicle driving. At present, a common target object detection scheme in automatic driving is BEV (Bird's-eye View) detection of laser radar point cloud (hereinafter referred to as point cloud), that is, three-dimensional point cloud is compressed into image data in the bird's-eye view, and then input into a 2D target detection algorithm for detection.

[0003] In the process of implementing the present application, the inventors found that the prior art has the following problems:

[0004] In the current mainstream point cloud feature extraction algorithm, it is usually necessary to perform voxelization on the point cloud based on the perspective of the bird's-eye view, so that the height information of the Z axis is greatly lost, and for different objects, the features at different heights have a great guiding effect on the identification task of the object. Therefore, at present, due to the loss of Z-axis height information, the feature extraction is incomplete, which greatly affects the accuracy of target object detection. SUMMARY

[0005] Therefore, the embodiments of the present application provide a target object detection method and device, which respectively perform first coordinate transformation and second coordinate transformation on point cloud points of original point cloud data to obtain first local features with spatial position information and second local features with height position information, and fuse the two local features and learn the neural network to obtain global features, so as to obtain a detection result according to a target detection model. Through the two times of coordinate transformation on the point cloud points, the extracted features include both the spatial position information and the height position information, which improves the accuracy of target object detection and better identifies the target object.

[0006] To achieve the object, according to an aspect of an embodiment of the present application, a target object detection method is provided, comprising: performing first coordinate transformation on each point cloud point of original point cloud data, and performing first feature extraction to obtain first local features, the first local features comprising spatial position information; performing second coordinate transformation on the point cloud point, and performing second feature extraction to obtain second local features, the second local features comprising height position information; fusing the first local features and the second local features to obtain target local features of the point cloud point; performing multi-layer perception learning on the target local features by using a neural network to obtain global features of the point cloud point; and inputting the global features into a target detection model to obtain a detection result of the target object.

[0007] Optionally, the first coordinate transformation on the point cloud point and the first feature extraction to obtain the first local features comprise: establishing a first transformation coordinate system, and performing voxel grid division on a point cloud space under the first transformation coordinate system; calculating first position information of the point cloud point under the first transformation coordinate system according to position information of the point cloud point; determining a voxel grid to which the point cloud point belongs according to the first position information, and calculating a deviation of the first position information from a center point of point distribution in the voxel grid to which the point cloud point belongs to obtain first deviation information; and splicing the first position information and the first deviation information to obtain the first local features of the point cloud point.

[0008] Optionally, the first local features further comprise intensity features of the point cloud point and number features of point cloud points included in the voxel grid to which the point cloud point belongs.

[0009] Optionally, the second coordinate transformation on the point cloud point and the second feature extraction to obtain the second local features comprise: establishing a second transformation coordinate system, and performing voxel grid division on a point cloud space under the second transformation coordinate system, the voxel grid being parallel to the ground; calculating second position information of the point cloud point under the second transformation coordinate system according to position information of the point cloud point; determining a voxel grid to which the point cloud point belongs according to the second position information, and calculating a deviation of the second position information from a center point of point distribution in the voxel grid to which the point cloud point belongs to obtain second deviation information; and splicing the second position information and the second deviation information to obtain the second local features.

[0010] Optionally, the second local features further comprise number features of point cloud points included in the voxel grid to which the point cloud point belongs.

[0011] Optionally, the center point of point distribution in the voxel grid to which the point cloud point belongs is determined according to an average value of position information of all point cloud points in the voxel grid to which the point cloud point belongs.

[0012] Optionally, the first coordinate transformation is to transform the position information of the point cloud point in an original Cartesian coordinate system into first position information in a target Cartesian coordinate system, and the second coordinate transformation is to transform the position information of the point cloud point in the original Cartesian coordinate system into second position information in a cylindrical coordinate system.

[0013] According to a second aspect of the embodiments of the present application, a detection device for a target object is provided, comprising: a first feature extraction module configured to perform first coordinate transformation on each point cloud point of original point cloud data, and perform first feature extraction to obtain a first local feature, the first local feature comprising spatial position information; a second feature extraction module configured to perform second coordinate transformation on the point cloud point, and perform second feature extraction to obtain a second local feature, the second local feature comprising height position information; a target local feature acquisition module configured to fuse the first local feature and the second local feature to obtain a target local feature of the point cloud point; a global feature acquisition module configured to perform multi-layer perception learning on the target local feature by using a neural network to obtain a global feature of the point cloud point; and a detection module configured to input the global feature into a target detection model to obtain a detection result of the target object.

[0014] According to a third aspect of the embodiments of the present application, a detection electronic device for a target object is provided, comprising:

[0015] one or more processors;

[0016] a storage device configured to store one or more programs,

[0017] when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided by the first aspect of the embodiments of the present application.

[0018] According to a fourth aspect of the embodiments of the present application, a computer readable medium having a computer program stored thereon is provided, the program being executed by a processor to implement the method provided by the first aspect of the embodiments of the present application.

[0019] An embodiment of the present application has the following advantages or beneficial effects: by performing a first coordinate transformation on each point cloud point of the original point cloud data, and performing first feature extraction to obtain first local features, the first local features include spatial position information; by performing a second coordinate transformation on the point cloud point, and performing second feature extraction to obtain second local features, the second local features include height position information; by fusing the first local features and the second local features, target local features of the point cloud point are obtained; by using a neural network to perform multi-layer perception learning on the target local features, global features of the point cloud point are obtained; by inputting the global features into a target detection model, a detection result of a target object is obtained, which realizes that by performing twice coordinate transformation on the point cloud point, the target local features extracted include both spatial position information and height position information, and then based on the neural network, learning is performed on the target local features to obtain global features for target object detection, thereby solving the problem of low accuracy of target object detection caused by the lack of height information of the point cloud during feature extraction in the prior art, and improving the accuracy of target object detection and better identifying the target object. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings are included to provide a further understanding of the present application, and are incorporated in and constitute a part of this specification. Among other things, the drawings illustrate embodiments of the present application.

[0021] Figure 1 is a schematic diagram of the main flow of the target object detection method according to an embodiment of the present application;

[0022] Figure 2 is a schematic diagram of the principle of an embodiment of the present application;

[0023] Figure 3 is a schematic diagram of the main modules of the target object detection device according to an embodiment of the present application;

[0024] Figure 4 is an exemplary system architecture diagram to which an embodiment of the present application can be applied;

[0025] Figure 5 is a structural schematic diagram of a computer system of a terminal device or a server suitable for implementing an embodiment of the present application. DETAILED DESCRIPTION

[0026] Exemplary embodiments of the present application are described below with reference to the accompanying drawings, which include various details of the embodiments of the present application to assist in understanding, and should be considered as merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. Also, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0027] Currently, mainstream point cloud feature extraction algorithms typically require voxelization of the point cloud based on a top-down view, which results in a significant loss of Z-axis height information. However, for different objects, features at different heights are highly instructive for object recognition tasks. Therefore, Z-axis height information is of great significance for improving the detection performance of target objects.

[0028] To address the aforementioned problems in existing technologies, this invention proposes a target object detection method. By performing two coordinate transformations on point cloud points, the extracted target local features include both spatial location information and height location information. Then, based on a neural network, the target local features are learned to obtain global features for target object detection. This solves the problem of low target object detection accuracy caused by the lack of height information in point clouds during feature extraction in existing technologies, thereby improving the accuracy of target object detection and better identifying target objects.

[0029] In the description of the embodiments of the present invention, the terms involved and their meanings are as follows:

[0030] Point cloud: In reverse engineering, the collection of point data on the surface of a product obtained through measuring instruments is also called a point cloud;

[0031] BEV: Bird's-eye View, a three-dimensional drawing created by looking down at the undulations of the ground from a high point using the principles of perspective and the high-viewpoint perspective method;

[0032] Cylindrical view: A view that retains information about the height of an object;

[0033] Voxelization: Converting the geometric representation of an object into a voxel representation that is closest to the object, producing a volume dataset that not only contains surface information of the model but also describes the model's internal properties.

[0034] MLP: multi-layer perceptron, a type of feedforward artificial neural network model that maps multiple input datasets to a single output dataset.

[0035] Figure 1 This is a schematic diagram of the main flow of the target object detection method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the target object detection method of this embodiment of the invention includes the following steps S101 to S105.

[0036] Step S101: For each point in the original point cloud data, perform a first coordinate transformation on the point cloud point and extract a first feature to obtain a first local feature, wherein the first local feature includes spatial location information.

[0037] According to one embodiment of the present application, the first coordinate transformation is to transform the position information of the point cloud points in the original Cartesian coordinate system to the first position information in the target Cartesian coordinate system.

[0038] Specifically, in the field of automatic driving, the category and position information of obstacles are determined through laser radar point cloud detection, so as to make reasonable avoidance plan for different obstacles according to the detection results, to ensure the safety of vehicle driving. The point cloud can reflect the shape and posture information of the target object, but lacks texture information, therefore, in order to realize the detection of 3D target object, feature extraction needs to be performed on the point cloud data. Based on the original point cloud data, the center point of the point cloud distribution is taken as the origin in the BEV (Bird's-eye View) Cartesian coordinate system, in order to facilitate the detection of the target object, the original Cartesian coordinates of the point cloud points need to be transformed to the target Cartesian coordinate system, so that the position information of the point cloud points is positive in the target Cartesian coordinate system.

[0039] According to another embodiment of the present application, the first coordinate transformation is performed on the point cloud points, and the first local feature is obtained through first feature extraction, comprising: establishing a first transformation coordinate system, and performing voxel grid division on the point cloud space in the first transformation coordinate system; calculating the first position information of the point cloud points in the first transformation coordinate system according to the position information of the point cloud points; determining the voxel grid to which the point cloud points belong according to the first position information, and calculating the deviation of the first position information from the center point of the point distribution in the voxel grid to which the point cloud points belong, to obtain the first deviation information; splicing the first position information and the first deviation information to obtain the first local feature of the point cloud points.

[0040] According to still another embodiment of the present application, the first local feature further comprises the intensity feature of the point cloud points and the number feature of the point cloud points included in the voxel grid to which the point cloud points belong.

[0041] Specifically, according to the BEV Cartesian coordinate system of the original point cloud points, a target Cartesian coordinate system is established which satisfies that the position information of the point cloud points is positive. Exemplarily, for the point cloud points scanned by the laser radar, the origin of the original Cartesian coordinate system is moved to the lower left according to the distribution characteristics of the point cloud points, and the coordinate system which satisfies that the position information of the point cloud points is positive is the target Cartesian coordinate system. Based on the target Cartesian coordinate system, the point cloud space is divided into voxel grids, exemplarily, the voxel grid is HxWx1, wherein H and W are set according to the requirements of target object detection, and can be expressed as C bev =U i H*W*1 , which represents that U is C beva voxel grid; since the target Cartesian coordinate system is obtained based on the original Cartesian coordinate system, the position information of the original point cloud point is translated to obtain the first position information, and the information of a point of the original point cloud is (x, y, z, r), where (x, y, z) is the position information, and r is the reflection intensity of the current point, and after the first transformation, the first position information of the current point is (x1, y1, z), where x1 and y1 are the position information after left and downward translation respectively; according to the position information of the point cloud point transformed into the target Cartesian coordinate system, and in combination with the division of the voxel grid, the voxel grid in which the point cloud point is located is determined, the number of point cloud points in the voxel grid is obtained, for the distribution of the point cloud in the voxel grid, the center point of the point cloud distribution is found, and the deviation of the position information of the point in the voxel grid from the center point is calculated, that is, the first deviation information; the first position information and the first deviation information are spliced, and since the reflection intensity of the current point and the number of point cloud points in the grid are also the feature information of the point cloud point, the above position information, deviation information, reflection intensity information and number information are combined to obtain the first local feature of the point cloud point.

[0042] For example, it is assumed that the information of a point of the original point cloud is (x, y, z, r), the first position information of the current point in the target Cartesian coordinate is (x1, y1, z, r), the first deviation information in the voxel grid U is (x', y', z'), the number of point cloud points in the voxel grid is n, and finally the above information is combined to obtain the first local feature P of the current point. u u u u bev

[0043] bev u u u u bev .

[0044] According to another embodiment of the present application, the center point of the point distribution in the voxel grid to which the point cloud point belongs is determined according to the average value of the position information of all point cloud points in the voxel grid to which the point cloud point belongs.

[0045] For example, it is assumed that the position information of each point cloud point in the voxel grid is (x i , y i , z i ), and there are N point cloud points, and then the center point of the voxel grid is the arithmetic average value of the position information of the point cloud points ​​​​​​​​​​​

[0046] The first coordinate transformation in the Cartesian coordinate system from the perspective of the BEV and the voxelization feature extraction of the point cloud points ensure that the position information of the object in space is retained in the feature extraction information.

[0047] In step S102, the second coordinate transformation is performed on the point cloud points, and the second local feature is obtained through the second feature extraction. The second local feature includes the height position information.

[0048] According to an embodiment of the present application, the second coordinate transformation is to transform the position information of the point cloud points in the original Cartesian coordinate system into the second position information in the cylindrical coordinate system.

[0049] Specifically, according to the BEV Cartesian coordinate system of the original point cloud points, the position information of the point cloud points in the original Cartesian coordinate system is subjected to the second coordinate transformation in the cylindrical coordinate system, to obtain the second position information of the original point cloud points after the second coordinate transformation. The feature extraction is performed based on the second position information, to obtain the second local feature. Since the cylindrical coordinate in the cylindrical perspective retains the height information of the target object, the feature information of the target object can be enriched, to improve the accuracy of the target object detection.

[0050] According to another embodiment of the present application, the second coordinate transformation is performed on the point cloud points, and the second local feature is obtained through the second feature extraction, including: establishing a second transformation coordinate system, and performing the voxel grid division on the point cloud space in the second transformation coordinate system. The voxel grid is parallel to the ground; calculating the second position information of the point cloud points in the second transformation coordinate according to the position information of the point cloud points; determining the voxel grid to which the point cloud points belong according to the second position information, and calculating the deviation of the second position information from the center point of the point distribution in the voxel grid to which the point cloud points belong, to obtain the second deviation information; and splicing the second position information and the second deviation information, to obtain the second local feature.

[0051] According to still another embodiment of the present application, the second local feature further includes the quantity feature of the point cloud points included in the voxel grid to which the point cloud points belong.

[0052] Specifically, the cylindrical coordinate system is established, taking the radar as the center, and projecting the voxel to the periphery, so that a plurality of voxel grids parallel to the ground are formed around the periphery, to realize the voxel grid division of the point cloud space; the second position information converted from the original Cartesian coordinate system to the cylindrical coordinate system is calculated according to the position information of the original point cloud points. Exemplarily, p i (x i ,y i ,z i ) is the position information of the original point cloud points in the original Cartesian coordinate system, and p iCorresponding coordinates in a cylindrical coordinate system are:

[0053]

[0054] According to the second position information, in combination with the division of the voxel grid, the voxel grid to which the point cloud point belongs is determined, the number of point cloud points in the voxel grid is obtained, for the distribution of the point cloud in the voxel grid, the center point of the point cloud distribution is found, the deviation of the position information of the points in the voxel grid from the center point is calculated, that is, the second deviation information; the second position information and the second deviation information are spliced, since the number of point cloud points in the grid is also the feature information of the point cloud points, therefore, the above position information, deviation information and number information are combined to obtain the second local feature of the point cloud point.

[0055] Exemplarily, the information of an original point cloud point of a point is set as (x i , y i , z i ), the second position information of the current point in the cylindrical coordinate is The second deviation information in the voxel grid U is The number of point cloud points in the voxel grid is n u_cyu Finally, the above information is combined to obtain the second local feature P clinder of the current point.

[0056]

[0057] According to another embodiment of the application, the center point of the point distribution in the voxel grid to which the point cloud point belongs is determined according to the average value of the position information of all point cloud points in the voxel grid to which the point cloud point belongs.

[0058] Generally, the method of determining the center point is similar to that of the first coordinate transformation, which will not be repeated here.

[0059] Through the above second transformation from the original Cartesian coordinate system to the cylindrical coordinate system of the cylindrical view angle, and the voxelization feature extraction of the point cloud point, it is ensured that the height information of the object is retained in the feature extraction information, and at the same time, the view angle conforms to the imaging principle of the radar, and can more accurately represent the features of the radar imaging.

[0060] Step S103, fusing the first local feature and the second local feature to obtain the target local feature of the point cloud point.

[0061] Specifically, according to the first transformation of the BEV Cartesian coordinate system and the second transformation of the cylindrical coordinate system of the cylindrical view angle, the feature extraction based on the spatial position information of the first transformation and the feature extraction based on the height position information of the second transformation are ensured, the feature values of the two coordinate systems are fused, the features of the two view angles are complementary, the target local feature of the point cloud point obtained includes the spatial position information and the height position information, and the target local feature of the point cloud point is exemplified as P f For:

[0062]

[0063] Step S104, the target local feature is subjected to multi-layer perception learning by using a neural network, and the global feature of the point cloud point is obtained.

[0064] Specifically, the target local feature of the point cloud point is taken as an input of MLP (mutil layer perceptron), the neural network is used to perform multi-layer perception learning on the target local feature, and the global feature of the point cloud point is obtained, so that subsequent target object detection is performed.

[0065] Step S105, the global feature is input into a target detection model, and a detection result of a target object is obtained.

[0066] Specifically, the global feature is input into the target detection model, the category and position information of the target object are obtained by using a target detection algorithm, and a reasonable avoidance plan can be made according to the information of the obstacle.

[0067] Figure 2 is a principle schematic diagram of an embodiment of the present application, in which the original point cloud data is subjected to spatial position information feature extraction of point cloud points by bird's eye view angle voxelization and height position information feature extraction of point cloud points by cylindrical view angle voxelization, so as to obtain first local features of a Cartesian coordinate system of a bird's eye view angle and second local features of a cylindrical coordinate system of a cylindrical view angle; the point-level features of the bird's eye view angle and the point-level features of the cylindrical view angle are fused to obtain target local feature information; finally, the global feature information of the point cloud point is obtained through a multi-layer perceptron, and a detection result of a target object is obtained through a detection model.

[0068] Figure 3 is a main module schematic diagram of a target object detection device according to an embodiment of the present application. As shown in Figure 3 the target object detection device 300 mainly includes a first feature extraction module 301, a second feature extraction module 302, a target local feature acquisition module 303, a global feature acquisition module 304, and a detection module 305.

[0069] The first feature extraction module 301 is configured to perform first coordinate transformation on each point cloud point of the original point cloud data, and perform first feature extraction to obtain a first local feature of the point cloud point, the first local feature comprising spatial position information.

[0070] The second feature extraction module 302 is configured to perform second coordinate transformation on the point cloud point, and perform second feature extraction to obtain a second local feature of the point cloud point, the second local feature comprising height position information.

[0071] The target local feature acquisition module 303 is configured to fuse the first local feature and the second local feature to obtain a target local feature of the point cloud point.

[0072] The global feature acquisition module 304 is configured to perform multi-layer perception learning on the target local feature by using a neural network to obtain a global feature of the point cloud point.

[0073] The detection module 305 is configured to input the global feature into a target detection model to obtain a detection result of a target object.

[0074] Specifically, the first feature extraction module 301 is further configured to establish a first transformation coordinate system, perform voxel grid division on a point cloud space in the first transformation coordinate system, calculate first position information of the point cloud point in the first transformation coordinate system according to position information of the point cloud point, determine a voxel grid to which the point cloud point belongs according to the first position information, calculate a deviation between the first position information and a center point of a point distribution in the voxel grid to which the point cloud point belongs to obtain first deviation information, and splice the first position information and the first deviation information to obtain the first local feature of the point cloud point.

[0075] Specifically, the first local feature further comprises intensity features of the point cloud point and quantity features of point cloud points included in the voxel grid to which the point cloud point belongs.

[0076] Specifically, the second feature extraction module 302 is further configured to establish a second transformation coordinate system, perform voxel grid division on a point cloud space in the second transformation coordinate system, the voxel grid being parallel to the ground, calculate second position information of the point cloud point in the second transformation coordinate system according to position information of the point cloud point, determine a voxel grid to which the point cloud point belongs according to the second position information, calculate a second deviation information between the second position information and a center point of a point distribution in the voxel grid to which the point cloud point belongs, and splice the second position information and the second deviation information to obtain the second local feature.

[0077] Specifically, the second local feature further comprises quantity features of point cloud points included in the voxel grid to which the point cloud point belongs.

[0078] Specifically, the center point of the point distribution within the voxel grid to which the point cloud point belongs is determined based on the average value of the position information of all point cloud points within the voxel grid to which the point cloud point belongs.

[0079] Specifically, the first coordinate transformation transforms the position information of the point cloud points in the original Cartesian coordinate system to the first position information in the target Cartesian coordinate system, and the second coordinate transformation transforms the position information of the point cloud points in the original Cartesian coordinate system to the second position information in the cylindrical coordinate system.

[0080] Figure 4 This is an exemplary system architecture diagram in which embodiments of the present invention can be applied.

[0081] like Figure 4 As shown, system architecture 400 may include terminal devices 401, 402, and 403, a network 404, and a server 405. Network 404 serves as the medium for providing communication links between terminal devices 401, 402, and 403 and server 405. Network 404 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0082] Users can use terminal devices 401, 402, and 403 to interact with server 405 via network 404 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 401, 402, and 403, such as target object detection applications, target object recognition applications, etc. (for example only).

[0083] Terminal devices 401, 402, and 403 can be various electronic devices with displays that support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0084] The server 405 can be a server providing various services, for example, a background management server providing support for detection of a target object by a user using a terminal device 401, 402, 403 (only as an example). The background management server can, for each point cloud point of raw point cloud data, perform first coordinate transformation on the point cloud point, and perform first feature extraction to obtain first local features, the first local features including spatial position information; perform second coordinate transformation on the point cloud point, and perform second feature extraction to obtain second local features, the second local features including height position information; fuse the first local features and the second local features to obtain target local features of the point cloud point; perform multi-layer perception learning on the target local features by using a neural network to obtain global features of the point cloud point; input the global features into a target detection model to obtain a detection result of a target object and the like, and feed back a processing result (for example, a detection result and the like--only as an example) to a terminal device.

[0085] It should be noted that the target object detection method provided by the embodiment of the present application is generally executed by the server 405, and correspondingly, the target object detection device is generally arranged in the server 405.

[0086] It should be understood that, Figure 4 The number of terminal devices, networks and servers in the system is only illustrative. According to the needs of implementation, there can be any number of terminal devices, networks and servers.

[0087] Reference will be made to Figure 5 which shows a structural schematic diagram of a computer system 500 suitable for implementing a terminal device or a server of the embodiment of the present application. Figure 5 The terminal device or server shown is only an example, and should not bring any limitation to the function and use range of the embodiment of the present application.

[0088] As shown in Figure 5 , the computer system 500 includes a central processing unit (CPU) 501 which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 502 or programs loaded from a storage portion 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the system 500 are also stored. The CPU 501, the ROM 502 and the RAM 503 are connected to each other through a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0089] The following components are connected to the I / O interface 505: an input part 506 including a keyboard, a mouse, etc.; an output part 507 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage part 508 including a hard disk, etc.; and a communication part 509 including a network interface card such as a LAN card, a modem, etc. The communication part 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the I / O interface 505 as necessary. A removable medium 511 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 510 as necessary, so that a computer program read out therefrom is installed in the storage part 508 as necessary.

[0090] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to embodiments of the present disclosure. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for executing the methods illustrated by the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network by the communication part 509, and / or installed from the removable medium 511. When the computer program is executed by the central processing unit (CPU) 501, the above-described functions defined in the system of the present disclosure are executed.

[0091] It should be noted that the computer-readable medium shown in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or component. In the present application, the computer-readable signal medium can include a data signal carried in a baseband or as a carrier wave part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take many forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, device or component. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0092] The flow diagrams and block diagrams in the drawings are schematic illustrations of possible architectures, functions and operations of systems, methods and computer program products in accordance with various embodiments of the present application. In this regard, each block in the flow diagrams or block diagrams can represent a module, a segment, or a portion of code, which comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks may, in fact, be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flow diagrams, and combinations of blocks in the block diagrams or flow diagrams, can be implemented by special purpose hardware-based systems that perform the specified functions or operations, or combinations of special purpose hardware and computer instructions.

[0093] The units described in the embodiments of the present application can be implemented in the form of software, or can be implemented in the form of hardware. The described units can also be arranged in a processor, for example, a processor can be described as: a processor comprising a first feature extraction module, a second feature extraction module, a target local feature acquisition module, a global feature acquisition module and a detection module.

[0094] In some cases, the names of the modules do not constitute a limitation of the modules themselves, for example, the detection module can also be described as: a module for inputting the global feature into a target detection model to obtain a detection result of a target object.

[0095] In another aspect, the present application also provides a computer readable medium, which can be included in the device described in the embodiments, or can exist independently without being assembled into the device. The computer readable medium carries one or more programs, when the one or more programs are executed by the device, the device comprises: for each point cloud point of the original point cloud data, performing first coordinate transformation on the point cloud point, and performing first feature extraction to obtain first local feature, the first local feature comprising spatial position information; performing second coordinate transformation on the point cloud point, and performing second feature extraction to obtain second local feature, the second local feature comprising height position information; fusing the first local feature and the second local feature to obtain target local feature of the point cloud point; using a neural network to perform multi-layer perception learning on the target local feature to obtain global feature of the point cloud point; inputting the global feature into a target detection model to obtain a detection result of a target object.

[0096] According to the technical scheme of the embodiment of the present application, the following advantages or beneficial effects are achieved: by performing first coordinate transformation on each point cloud point of the original point cloud data, and performing first feature extraction to obtain first local features, the first local features include spatial position information; by performing second coordinate transformation on the point cloud point, and performing second feature extraction to obtain second local features, the second local features include height position information; by fusing the first local features and the second local features, target local features of the point cloud point are obtained; by using a neural network to perform multi-layer perception learning on the target local features, global features of the point cloud point are obtained; by inputting the global features into a target detection model, a detection result of a target object is obtained; the technical scheme realizes that by performing twice coordinate transformation on the point cloud point, the target local features extracted include both spatial position information and height position information, and then based on the neural network, learning is performed on the target local features to obtain global features, so as to perform target object detection, thereby solving the problem of low accuracy of target object detection caused by the lack of height information of the point cloud during feature extraction in the prior art, and thereby improving the accuracy of target object detection and better identifying the target object.

[0097] The specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method of detecting a target object, characterized by, The method comprises: performing first coordinate transformation on each point cloud point of original point cloud data, and performing first feature extraction to obtain first local features, the first local features comprising spatial position information; the first coordinate transformation is to transform position information of the point cloud point in an original Cartesian coordinate system into first position information in a target Cartesian coordinate system; performing second coordinate transformation on the point cloud point, and performing second feature extraction to obtain second local features, the second local features comprising height position information; the second coordinate transformation is to transform position information of the point cloud point in the original Cartesian coordinate system into second position information in a cylindrical coordinate system; before transforming the position information of the point cloud point in the original Cartesian coordinate system into the second position information in the cylindrical coordinate system, a cylindrical coordinate system is established, and voxel projection is performed around the radar as the axis, and voxel grid division is performed on the point cloud space; fusing the first local features and the second local features to obtain target local features of the point cloud point; performing multi-layer perception learning on the target local features by using a neural network to obtain global features of the point cloud point; inputting the global features into a target detection model to obtain a detection result of a target object.

2. The method of claim 1, wherein, The first coordinate transformation on the point cloud point and the first feature extraction to obtain the first local features comprise: establishing a first transformation coordinate system, and performing voxel grid division on the point cloud space in the first transformation coordinate system; calculating first position information of the point cloud point in the first transformation coordinate system according to position information of the point cloud point; determining a voxel grid to which the point cloud point belongs according to the first position information, and calculating a deviation of the first position information from a center point of point distribution in the voxel grid to which the point cloud point belongs to obtain first deviation information; splicing the first position information and the first deviation information to obtain the first local features of the point cloud point.

3. The method of claim 2, wherein, The first local features further comprise intensity features of the point cloud point and quantity features of point cloud points included in the voxel grid to which the point cloud point belongs.

4. The method of claim 1, wherein, The second coordinate transformation on the point cloud point and the second feature extraction to obtain the second local features comprise: establishing a second transformation coordinate system, and performing voxel grid division on the point cloud space in the second transformation coordinate system, the voxel grid being parallel to the ground; calculating second position information of the point cloud point in the second transformation coordinate system according to position information of the point cloud point; determining a voxel grid to which the point cloud point belongs according to the second position information, and calculating a deviation of the second position information from a center point of point distribution in the voxel grid to which the point cloud point belongs to obtain second deviation information; splicing the second position information and the second deviation information to obtain the second local features.

5. The method of claim 4, wherein, The second local features further comprise quantity features of point cloud points included in the voxel grid to which the point cloud point belongs.

6. The method according to any one of claims 2-5, characterized in that, The center point of point distribution in the voxel grid to which the point cloud point belongs is determined according to an average value of position information of all point cloud points in the voxel grid to which the point cloud point belongs.

7. A device for detecting a target object, characterized by comprising: The method comprises: The first feature extraction module is configured to perform first coordinate transformation on each point cloud point of the original point cloud data, and perform first feature extraction to obtain first local features, wherein the first local features comprise spatial position information; the first coordinate transformation is to transform position information of the point cloud point in an original Cartesian coordinate system into first position information in a target Cartesian coordinate system; The second feature extraction module is configured to perform second coordinate transformation on the point cloud point, and perform second feature extraction to obtain second local features, wherein the second local features comprise height position information; the second coordinate transformation is to transform position information of the point cloud point in the original Cartesian coordinate system into second position information in a cylindrical coordinate system; before transforming the position information of the point cloud point in the original Cartesian coordinate system into the second position information in the cylindrical coordinate system, a cylindrical coordinate system is established, and the cylindrical coordinate system is projected to the surrounding voxels with the radar as the axis, and the point cloud space is divided into a voxel grid; The target local feature acquisition module is configured to fuse the first local features and the second local features to obtain target local features of the point cloud point; The global feature acquisition module is configured to perform multi-layer perception learning on the target local features by using a neural network to obtain global features of the point cloud point; The detection module is configured to input the global features into a target detection model to obtain a detection result of a target object.

8. A mobile electronic device terminal, characterized by The method comprises: one or more processors; a storage device configured to store one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1-6.

9. A computer readable medium having stored thereon a computer program, characterized in that The program is executed by the processor to implement the method according to any one of claims 1-6.

Images