Radar point cloud target classification method and system

Abstract

The invention provides a radar point cloud target classification method and system, and the method comprises the steps: carrying out the preprocessing of original radar point cloud data, so as to generate corresponding standard radar point cloud data; performing feature extraction processing on the standard radar point cloud data through a preset Mamba model to obtain corresponding point cloud feature information, and correspondingly inputting the point cloud feature information into a preset deep neural network to train a corresponding target detection model; when it is detected that the vehicle is in the driving state in real time, actual radar point cloud data and actual environment images corresponding to the vehicle are collected in real time, and fusion processing is carried out on the actual radar point cloud data and the actual environment images to generate corresponding fusion point cloud data; and correspondingly inputting the fused point cloud data into a target detection model to enable the target detection model to output a plurality of corresponding actual point cloud targets, and correspondingly completing classification processing of the plurality of actual point cloud targets. The working efficiency of the radar can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of automobile radar technology, and in particular to a radar point cloud target classification method and system. Background Art

[0002] With the advancement of science and technology and the rapid development of productivity, cars have become popular in people's daily lives and have become one of the indispensable means of transportation for people's daily travel, greatly facilitating people's lives.

[0003] Among them, with the rapid development of automobile autonomous driving technology, how to improve the vehicle's perception ability has become one of the important research directions of automobiles. Specifically, existing radar sensors have become important perception devices for autonomous driving vehicles because they can work stably in complex environments such as bad weather and night.

[0004] Furthermore, existing radars generate a large amount of data during actual operation, which greatly increases the data processing volume of the radar, resulting in poor real-time performance of radar detection. In addition, when existing radars process multiple targets at the same time, missed detections and false detections are prone to occur, which correspondingly reduces the detection accuracy and work efficiency of the radar. Summary of the invention

[0005] Based on this, the purpose of the present invention is to provide a radar point cloud target classification method and system to solve the problems of poor real-time detection in the prior art and easy missed detection and false detection.

[0006] The first aspect of the embodiment of the present invention proposes:

[0007] A radar point cloud target classification method, wherein the method comprises:

[0008] Receiving raw radar point cloud data collected by the radar in real time, and preprocessing the raw radar point cloud data to generate corresponding standard radar point cloud data;

[0009] Perform feature extraction processing on the standard radar point cloud data through a preset Mamba model to obtain corresponding point cloud feature information, and input the point cloud feature information into a preset deep neural network to train a corresponding target detection model;

[0010] When it is detected in real time that the vehicle is in a driving state, actual radar point cloud data and an actual environment image corresponding to the vehicle are collected in real time, and the actual radar point cloud data and the actual environment image are fused to generate corresponding fused point cloud data;

[0011] The fused point cloud data is input into the target detection model accordingly, so that the target detection model outputs a number of corresponding actual point cloud targets, and the classification processing of the number of actual point cloud targets is completed accordingly, and each of the actual point cloud targets is unique.

[0012] The beneficial effect of the present invention is that by acquiring the required standard radar point cloud data in real time, the interference caused by redundant data can be avoided. Based on this, the required target detection model is further trained. Furthermore, in order to effectively distinguish each point cloud target, it is necessary to further obtain fused point cloud data. On this basis, it is only necessary to input the current fused point cloud data into the constructed target detection model, so that each actual point cloud target can be output in real time and the corresponding classification processing can be completed, thereby correspondingly improving the working efficiency of the radar.

[0013] Furthermore, the step of performing feature extraction processing on the standard radar point cloud data by using a preset Mamba model to obtain corresponding point cloud feature information includes:

[0014] When the standard radar point cloud data is acquired in real time, point cloud spatial features corresponding to the standard radar point cloud data are matched in real time through the preset Mamba model;

[0015] Performing multi-layer convolution and pooling processing on the standard radar point cloud data based on the point cloud spatial features to extract local information and global information corresponding to the standard radar point cloud data;

[0016] Feature extraction processing is performed on the local information and the global information to obtain the corresponding point cloud feature information.

[0017] Furthermore, the step of performing feature extraction processing on the local information and the global information to obtain the point cloud feature information accordingly includes:

[0018] When the local information and the global information are respectively acquired, a point network corresponding to the standard radar point cloud data is matched in real time in a preset database;

[0019] The local information and the global information are subjected to feature extraction processing through the point network to obtain the corresponding point cloud feature information.

[0020] Furthermore, the step of performing feature extraction processing on the local information and the global information through the point network to obtain the point cloud feature information accordingly includes:

[0021] When the point network is acquired in real time, a number of target network nodes correspondingly included in the point network are detected in real time;

[0022] Passing the local information and the global information through each of the target network nodes in turn to output corresponding point cloud geometric features and point cloud topological structures;

[0023] The point cloud geometric features and the point cloud topological structure are integrated to generate the point cloud feature information accordingly.

[0024] Furthermore, the step of fusing the actual radar point cloud data and the actual environment image to generate corresponding fused point cloud data includes:

[0025] When the actual radar point cloud data and the actual environment image are respectively acquired, a first timestamp corresponding to the actual radar point cloud data and a second timestamp corresponding to the actual environment image are detected in real time;

[0026] In real time, a spatial calibration algorithm adapted to the actual radar point cloud data and the actual environment image is called out in a preset algorithm database;

[0027] Based on the first timestamp and the second timestamp, the actual radar point cloud data and the actual environment image are calibrated by the spatial calibration algorithm to generate corresponding matching point cloud data, and the fused point cloud data is generated according to the matching point cloud data.

[0028] Furthermore, the step of generating the fused point cloud data according to the matching point cloud data includes:

[0029] When the matching point cloud data is acquired in real time, the corresponding data fusion algorithm is called out in real time in the preset algorithm database;

[0030] The input layer, feature layer and decision layer respectively included in the data fusion algorithm are detected in real time, and the matching point cloud data are processed through the input layer, the feature layer and the decision layer to generate the corresponding fused point cloud data.

[0031] Furthermore, the step of processing the matching point cloud data through the input layer, the feature layer and the decision layer to generate the fused point cloud data accordingly includes:

[0032] Extracting a number of point cloud feature values ​​corresponding to the matching point cloud data in real time through the input layer, and performing feature fusion processing on the point cloud feature values ​​through the feature layer to generate a corresponding point cloud feature chain;

[0033] The point cloud feature chain is converted into corresponding target point cloud information through the decision layer, and the target point cloud information includes the fused point cloud data.

[0034] The second aspect of the embodiment of the present invention proposes:

[0035] A radar point cloud target classification system, wherein the system comprises:

[0036] A receiving module, used for receiving raw radar point cloud data collected by the radar in real time, and preprocessing the raw radar point cloud data to generate corresponding standard radar point cloud data;

[0037] An extraction module is used to perform feature extraction processing on the standard radar point cloud data through a preset Mamba model to obtain corresponding point cloud feature information, and input the point cloud feature information into a preset deep neural network to train a corresponding target detection model;

[0038] A fusion module, for collecting actual radar point cloud data and an actual environment image corresponding to the vehicle in real time when it is detected in real time that the vehicle is in a driving state, and fusing the actual radar point cloud data and the actual environment image to generate corresponding fused point cloud data;

[0039] The output module is used to input the fused point cloud data into the target detection model so that the target detection model outputs a number of corresponding actual point cloud targets and completes the classification processing of a number of the actual point cloud targets, and each of the actual point cloud targets is unique.

[0040] Furthermore, the extraction module is specifically used for:

[0041] When the standard radar point cloud data is acquired in real time, point cloud spatial features corresponding to the standard radar point cloud data are matched in real time through the preset Mamba model;

[0042] Performing multi-layer convolution and pooling processing on the standard radar point cloud data based on the point cloud spatial features to extract local information and global information corresponding to the standard radar point cloud data;

[0043] Feature extraction processing is performed on the local information and the global information to obtain the corresponding point cloud feature information.

[0044] Furthermore, the extraction module is specifically used for:

[0045] When the local information and the global information are respectively acquired, a point network corresponding to the standard radar point cloud data is matched in real time in a preset database;

[0046] The local information and the global information are subjected to feature extraction processing through the point network to obtain the corresponding point cloud feature information.

[0047] Furthermore, the extraction module is specifically used for:

[0048] When the point network is acquired in real time, a number of target network nodes correspondingly included in the point network are detected in real time;

[0049] Passing the local information and the global information through each of the target network nodes in turn to output corresponding point cloud geometric features and point cloud topological structures;

[0050] The point cloud geometric features and the point cloud topological structure are integrated to generate the point cloud feature information accordingly.

[0051] Furthermore, the fusion module is specifically used for:

[0052] When the actual radar point cloud data and the actual environment image are respectively acquired, a first timestamp corresponding to the actual radar point cloud data and a second timestamp corresponding to the actual environment image are detected in real time;

[0053] In real time, a spatial calibration algorithm adapted to the actual radar point cloud data and the actual environment image is called out in a preset algorithm database;

[0054] Based on the first timestamp and the second timestamp, the actual radar point cloud data and the actual environment image are calibrated by the spatial calibration algorithm to generate corresponding matching point cloud data, and the fused point cloud data is generated according to the matching point cloud data.

[0055] Furthermore, the fusion module is specifically used for:

[0056] When the matching point cloud data is acquired in real time, the corresponding data fusion algorithm is called out in real time in the preset algorithm database;

[0057] The input layer, feature layer and decision layer respectively included in the data fusion algorithm are detected in real time, and the matching point cloud data are processed through the input layer, the feature layer and the decision layer to generate the corresponding fused point cloud data.

[0058] Furthermore, the fusion module is specifically used for:

[0059] Extracting a number of point cloud feature values ​​corresponding to the matching point cloud data in real time through the input layer, and performing feature fusion processing on the point cloud feature values ​​through the feature layer to generate a corresponding point cloud feature chain;

[0060] The point cloud feature chain is converted into corresponding target point cloud information through the decision layer, and the target point cloud information includes the fused point cloud data.

[0061] The third aspect of the embodiment of the present invention proposes:

[0062] A computer comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the radar point cloud target classification method as described above when executing the computer program.

[0063] The fourth aspect of the embodiments of the present invention proposes:

[0064] A readable storage medium stores a computer program, wherein the program, when executed by a processor, implements the radar point cloud target classification method as described above.

[0065] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A flowchart of a radar point cloud target classification method provided by the first embodiment of the present invention;

[0067] Figure 2 This is a structural block diagram of a radar point cloud target classification system provided in the third embodiment of the present invention.

[0068] The following specific implementation manner will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0069] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Several embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0070] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0071] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0072] See also Figure 1 , shown is the radar point cloud target classification method provided by the first embodiment of the present invention. The radar point cloud target classification method provided by this embodiment can output each actual point cloud target in real time and complete the corresponding classification processing, thereby greatly improving the working efficiency of the radar.

[0073] Specifically, this embodiment provides:

[0074] A radar point cloud target classification method specifically comprises the following steps:

[0075] Step S10, receiving raw radar point cloud data collected by the radar in real time, and preprocessing the raw radar point cloud data to generate corresponding standard radar point cloud data;

[0076] Step S20, performing feature extraction processing on the standard radar point cloud data through a preset Mamba model to obtain corresponding point cloud feature information, and inputting the point cloud feature information into a preset deep neural network to train a corresponding target detection model;

[0077] Step S30, when it is detected in real time that the vehicle is in a driving state, actual radar point cloud data and an actual environment image corresponding to the vehicle are collected in real time, and the actual radar point cloud data and the actual environment image are fused to generate corresponding fused point cloud data;

[0078] Step S40, inputting the fused point cloud data into the target detection model accordingly, so that the target detection model outputs a number of corresponding actual point cloud targets, and correspondingly completes the classification processing of a number of the actual point cloud targets, each of which is unique.

[0079] Specifically, in this embodiment, it should be noted that, in order to objectively and accurately complete the classification of each radar point cloud target, it is necessary to obtain the corresponding information contained in each radar point cloud target. Based on this, after the server receives the original radar point cloud data collected by the radar in real time, it is necessary to immediately perform corresponding preprocessing on the current original radar point cloud data. Preferably, the current original radar point cloud data is preprocessed by filtering and noise reduction to remove invalid data and invalid noise. Based on this, the required standard radar point cloud data can be further obtained, and the current standard radar point cloud data will be further processed by the pre-set Mamba (language) model to extract the required point cloud feature information.

[0080] Furthermore, the present invention will further input the current point cloud feature information into the preset deep neural network, and can train the required target detection model accordingly, so as to complete the preliminary preparation work accordingly. Based on this, in the actual application process, when it is detected in real time that the vehicle is in a driving state, the actual radar point cloud data and the actual environment image corresponding to the current vehicle's driving environment will be immediately collected, and the current actual radar point cloud data and the actual environment image will be immediately fused to further fuse the required fused point cloud data. On this basis, it is only necessary to further input the current fused point cloud data into the above-mentioned target detection model, and a number of actual point cloud targets can be output accordingly, and finally the corresponding classification processing can be completed according to the type corresponding to each current actual point cloud target, thereby greatly improving the working efficiency of the radar.

[0081] Second embodiment

[0082] Furthermore, the step of performing feature extraction processing on the standard radar point cloud data by using a preset Mamba model to obtain corresponding point cloud feature information includes:

[0083] When the standard radar point cloud data is acquired in real time, point cloud spatial features corresponding to the standard radar point cloud data are matched in real time through the preset Mamba model;

[0084] Performing multi-layer convolution and pooling processing on the standard radar point cloud data based on the point cloud spatial features to extract local information and global information corresponding to the standard radar point cloud data;

[0085] Feature extraction processing is performed on the local information and the global information to obtain the corresponding point cloud feature information.

[0086] Furthermore, the step of performing feature extraction processing on the local information and the global information to obtain the point cloud feature information accordingly includes:

[0087] When the local information and the global information are respectively acquired, a point network corresponding to the standard radar point cloud data is matched in real time in a preset database;

[0088] The local information and the global information are subjected to feature extraction processing through the point network to obtain the corresponding point cloud feature information.

[0089] Furthermore, the step of performing feature extraction processing on the local information and the global information through the point network to obtain the point cloud feature information accordingly includes:

[0090] When the point network is acquired in real time, a number of target network nodes correspondingly included in the point network are detected in real time;

[0091] Passing the local information and the global information through each of the target network nodes in turn to output corresponding point cloud geometric features and point cloud topological structures;

[0092] The point cloud geometric features and the point cloud topological structure are integrated to generate the point cloud feature information accordingly.

[0093] Furthermore, the step of fusing the actual radar point cloud data and the actual environment image to generate corresponding fused point cloud data includes:

[0094] When the actual radar point cloud data and the actual environment image are respectively acquired, a first timestamp corresponding to the actual radar point cloud data and a second timestamp corresponding to the actual environment image are detected in real time;

[0095] In real time, a spatial calibration algorithm adapted to the actual radar point cloud data and the actual environment image is called out in a preset algorithm database;

[0096] Based on the first timestamp and the second timestamp, the actual radar point cloud data and the actual environment image are calibrated by the spatial calibration algorithm to generate corresponding matching point cloud data, and the fused point cloud data is generated according to the matching point cloud data.

[0097] Furthermore, the step of generating the fused point cloud data according to the matching point cloud data includes:

[0098] When the matching point cloud data is acquired in real time, the corresponding data fusion algorithm is called out in real time in the preset algorithm database;

[0099] The input layer, feature layer and decision layer respectively included in the data fusion algorithm are detected in real time, and the matching point cloud data are processed through the input layer, the feature layer and the decision layer to generate the corresponding fused point cloud data.

[0100] Furthermore, the step of processing the matching point cloud data through the input layer, the feature layer and the decision layer to generate the fused point cloud data accordingly includes:

[0101] Extracting a number of point cloud feature values ​​corresponding to the matching point cloud data in real time through the input layer, and performing feature fusion processing on the point cloud feature values ​​through the feature layer to generate a corresponding point cloud feature chain;

[0102] The point cloud feature chain is converted into corresponding target point cloud information through the decision layer, and the target point cloud information includes the fused point cloud data.

[0103] In addition, in this embodiment, it is also necessary to explain that after the required standard radar point cloud data is obtained in real time through the above steps, in order to objectively and accurately complete the corresponding feature extraction processing, preferably, the present invention will first match the corresponding point cloud spatial features through the above Mamba model, and use the current point cloud spatial features as the corresponding extraction target, perform multi-layer convolution and pooling processing on the current standard radar point cloud data, and obtain the required local information and global information accordingly. Further, in order to facilitate subsequent processing, it is necessary to further match the point network corresponding to the current standard radar point cloud data in real time in a pre-set database, and detect the corresponding number of target network nodes contained in the current electrical network. Preferably, the current local information and global information are immediately passed through each current target network node in turn, so that the required point cloud geometric features and point cloud topological structure can be further obtained. At the same time, corresponding integration processing is performed to obtain the above point cloud feature information.

[0104] Furthermore, in order to be able to quickly and accurately obtain the required fused point cloud data, specifically, the present invention will further detect the first timestamp corresponding to the above-mentioned actual radar point cloud data and the second timestamp corresponding to the above-mentioned actual environment image. At the same time, the adaptive spatial calibration algorithm is further called out, and the calibration process between the two is immediately completed through the spatial calibration algorithm, and the required matching point cloud data can be obtained. Furthermore, the data fusion algorithm adapted to the current matching point cloud data will be further called out, and the input layer, feature layer and decision layer contained in the current data fusion algorithm will be correspondingly detected. Preferably, several point cloud feature values ​​further contained in the current matching point cloud data are extracted in real time through the current input layer. Based on this, the corresponding feature fusion process is further performed on the current several point cloud feature values ​​through the above-mentioned feature layer to generate a corresponding point cloud feature chain. On this basis, the current point cloud feature chain is finally converted into the required target point cloud information through the decision layer, and the target point cloud information contains the required fused point cloud data, and the subsequent classification process is further completed, so that the working efficiency of the radar can be greatly improved.

[0105] See also Figure 2 , the third embodiment of the present invention provides:

[0106] A radar point cloud target classification system, wherein the system comprises:

[0107] A receiving module, used for receiving raw radar point cloud data collected by the radar in real time, and preprocessing the raw radar point cloud data to generate corresponding standard radar point cloud data;

[0108] An extraction module is used to perform feature extraction processing on the standard radar point cloud data through a preset Mamba model to obtain corresponding point cloud feature information, and input the point cloud feature information into a preset deep neural network to train a corresponding target detection model;

[0109] A fusion module, for collecting actual radar point cloud data and an actual environment image corresponding to the vehicle in real time when it is detected in real time that the vehicle is in a driving state, and fusing the actual radar point cloud data and the actual environment image to generate corresponding fused point cloud data;

[0110] The output module is used to input the fused point cloud data into the target detection model so that the target detection model outputs a number of corresponding actual point cloud targets and completes the classification processing of a number of the actual point cloud targets, and each of the actual point cloud targets is unique.

[0111] Furthermore, the extraction module is specifically used for:

[0112] When the standard radar point cloud data is acquired in real time, point cloud spatial features corresponding to the standard radar point cloud data are matched in real time through the preset Mamba model;

[0113] Performing multi-layer convolution and pooling processing on the standard radar point cloud data based on the point cloud spatial features to extract local information and global information corresponding to the standard radar point cloud data;

[0114] Feature extraction processing is performed on the local information and the global information to obtain the corresponding point cloud feature information.

[0115] Furthermore, the extraction module is specifically used for:

[0116] When the local information and the global information are respectively acquired, a point network corresponding to the standard radar point cloud data is matched in real time in a preset database;

[0117] The local information and the global information are subjected to feature extraction processing through the point network to obtain the corresponding point cloud feature information.

[0118] Furthermore, the extraction module is specifically used for:

[0119] When the point network is acquired in real time, a number of target network nodes correspondingly included in the point network are detected in real time;

[0120] Passing the local information and the global information through each of the target network nodes in turn to output corresponding point cloud geometric features and point cloud topological structures;

[0121] The point cloud geometric features and the point cloud topological structure are integrated to generate the point cloud feature information accordingly.

[0122] Furthermore, the fusion module is specifically used for:

[0123] When the actual radar point cloud data and the actual environment image are respectively acquired, a first timestamp corresponding to the actual radar point cloud data and a second timestamp corresponding to the actual environment image are detected in real time;

[0124] In real time, a spatial calibration algorithm adapted to the actual radar point cloud data and the actual environment image is called out in a preset algorithm database;

[0125] Based on the first timestamp and the second timestamp, the actual radar point cloud data and the actual environment image are calibrated by the spatial calibration algorithm to generate corresponding matching point cloud data, and the fused point cloud data is generated according to the matching point cloud data.

[0126] Furthermore, the fusion module is specifically used for:

[0127] When the matching point cloud data is acquired in real time, the corresponding data fusion algorithm is called out in real time in the preset algorithm database;

[0128] The input layer, feature layer and decision layer respectively included in the data fusion algorithm are detected in real time, and the matching point cloud data are processed through the input layer, the feature layer and the decision layer to generate the corresponding fused point cloud data.

[0129] Furthermore, the fusion module is specifically used for:

[0130] Extracting a number of point cloud feature values ​​corresponding to the matching point cloud data in real time through the input layer, and performing feature fusion processing on the point cloud feature values ​​through the feature layer to generate a corresponding point cloud feature chain;

[0131] The point cloud feature chain is converted into corresponding target point cloud information through the decision layer, and the target point cloud information includes the fused point cloud data.

[0132] A fourth embodiment of the present invention provides a computer, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the radar point cloud target classification method as described above when executing the computer program.

[0133] A fifth embodiment of the present invention provides a readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the radar point cloud target classification method as described above is implemented.

[0134] In summary, the radar point cloud target classification method and system provided by the above embodiments of the present invention can output each actual point cloud target in real time and complete the corresponding classification processing, thereby correspondingly improving the working efficiency of the radar.

[0135] It should be noted that the above modules can be functional modules or program modules, and can be implemented by software or hardware. For modules implemented by hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0136] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0137] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0138] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0139] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0140] The above-described embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the attached claims.

Claims

1. A radar point cloud target classification method, characterized in that: The method comprises: Receiving raw radar point cloud data collected by the radar in real time, and preprocessing the raw radar point cloud data to generate corresponding standard radar point cloud data; Perform feature extraction processing on the standard radar point cloud data through a preset Mamba model to obtain corresponding point cloud feature information, and input the point cloud feature information into a preset deep neural network to train a corresponding target detection model; When it is detected in real time that the vehicle is in a driving state, actual radar point cloud data and an actual environment image corresponding to the vehicle are collected in real time, and the actual radar point cloud data and the actual environment image are fused to generate corresponding fused point cloud data; The fused point cloud data is input into the target detection model accordingly, so that the target detection model outputs a number of corresponding actual point cloud targets, and the classification processing of the number of actual point cloud targets is completed accordingly, and each of the actual point cloud targets is unique.

2. The radar point cloud target classification method according to claim 1, characterized in that: The step of performing feature extraction processing on the standard radar point cloud data by using a preset Mamba model to obtain corresponding point cloud feature information comprises: When the standard radar point cloud data is acquired in real time, point cloud spatial features corresponding to the standard radar point cloud data are matched in real time through the preset Mamba model; Performing multi-layer convolution and pooling processing on the standard radar point cloud data based on the point cloud spatial features to extract local information and global information corresponding to the standard radar point cloud data; Feature extraction processing is performed on the local information and the global information to obtain the corresponding point cloud feature information.

3. The radar point cloud target classification method according to claim 2, characterized in that: The step of performing feature extraction processing on the local information and the global information to obtain the point cloud feature information accordingly includes: When the local information and the global information are respectively acquired, a point network corresponding to the standard radar point cloud data is matched in real time in a preset database; The local information and the global information are subjected to feature extraction processing through the point network to obtain the corresponding point cloud feature information.

4. The radar point cloud target classification method according to claim 3, characterized in that: The step of performing feature extraction processing on the local information and the global information through the point network to obtain the point cloud feature information accordingly includes: When the point network is acquired in real time, a number of target network nodes correspondingly included in the point network are detected in real time; Passing the local information and the global information through each of the target network nodes in sequence to output corresponding point cloud geometric features and point cloud topological structures; The point cloud geometric features and the point cloud topological structure are integrated to generate the point cloud feature information accordingly.

5. The radar point cloud target classification method according to claim 1, characterized in that: The step of fusing the actual radar point cloud data and the actual environment image to generate corresponding fused point cloud data includes: When the actual radar point cloud data and the actual environment image are respectively acquired, a first timestamp corresponding to the actual radar point cloud data and a second timestamp corresponding to the actual environment image are detected in real time; In real time, a spatial calibration algorithm adapted to the actual radar point cloud data and the actual environment image is called out in a preset algorithm database; Based on the first timestamp and the second timestamp, the actual radar point cloud data and the actual environment image are calibrated by the spatial calibration algorithm to generate corresponding matching point cloud data, and the fused point cloud data is generated according to the matching point cloud data.

6. The radar point cloud target classification method according to claim 5, characterized in that: The step of generating the fused point cloud data according to the matching point cloud data comprises: When the matching point cloud data is acquired in real time, the corresponding data fusion algorithm is called out in real time in the preset algorithm database; The input layer, feature layer and decision layer respectively included in the data fusion algorithm are detected in real time, and the matching point cloud data are processed through the input layer, the feature layer and the decision layer to generate the corresponding fused point cloud data.

7. The radar point cloud target classification method according to claim 6, characterized in that: The step of processing the matching point cloud data through the input layer, the feature layer and the decision layer to generate the fused point cloud data accordingly includes: Extracting a number of point cloud feature values ​​corresponding to the matching point cloud data in real time through the input layer, and performing feature fusion processing on the point cloud feature values ​​through the feature layer to generate a corresponding point cloud feature chain; The point cloud feature chain is converted into corresponding target point cloud information through the decision layer, and the target point cloud information includes the fused point cloud data.

8. A radar point cloud target classification system, characterized in that: The system comprises: A receiving module, used for receiving raw radar point cloud data collected by the radar in real time, and preprocessing the raw radar point cloud data to generate corresponding standard radar point cloud data; An extraction module is used to perform feature extraction processing on the standard radar point cloud data through a preset Mamba model to obtain corresponding point cloud feature information, and input the point cloud feature information into a preset deep neural network to train a corresponding target detection model; A fusion module, for collecting actual radar point cloud data and actual environment images corresponding to the vehicle in real time when it is detected in real time that the vehicle is in a driving state, and fusing the actual radar point cloud data and the actual environment images to generate corresponding fused point cloud data; The output module is used to input the fused point cloud data into the target detection model so that the target detection model outputs a number of corresponding actual point cloud targets and completes the classification processing of a number of the actual point cloud targets, and each of the actual point cloud targets is unique.

9. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the radar point cloud target classification method according to any one of claims 1 to 7 is implemented.

10. A readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the radar point cloud target classification method as described in any one of claims 1 to 7 is implemented.

Images