Obstacle detection method, device and electronic equipment

Abstract

The present invention provides an obstacle detection method, device, and electronic device, comprising collecting radar point cloud data; extracting feature information of each radar point from the radar point cloud data; inputting the feature information of each radar point into an obstacle detection model, and outputting an obstacle detection result; wherein the obstacle detection model is trained using radar point cloud data samples corresponding to at least one labeled obstacle. The obstacle detection method and related equipment provided by the present invention extract features from the radar point cloud to obtain information in multiple dimensions of the radar point cloud, input the feature information in multiple dimensions into a neural network model, and detect and identify obstacle types using the neural network model. The present invention has low reliance on pre-processing and other pre-processing processes. By processing the multi-dimensional features of the radar point cloud through a neural network model, not only is the process relatively simple and the speed relatively fast, but the stability and accuracy of obstacle detection and identification are also improved.

Description

Technical Field

[0001] The present invention relates to the field of radar technology, and in particular to an obstacle detection method, device and electronic equipment. Background Art

[0002] Millimeter-wave radar operates in the millimeter-wave band, which spans 30-300 GHz and has a wavelength of 1-10 mm. Millimeter-wave radar can identify very small targets and multiple targets simultaneously. It boasts strong imaging capabilities, a compact size, excellent maneuverability, and stealth. Due to its lack of blind spots and ease of miniaturization, millimeter-wave radar is widely used in the automotive sector, becoming a crucial component for sensing and detecting its surroundings.

[0003] Traditional obstacle detection methods manually set conditions and thresholds on the point cloud generated by millimeter-wave radar points, remove noise and interference, and then use the DBSCAN clustering algorithm to detect targets. However, automotive millimeter-wave radar point clouds have multiple information dimensions and are sparse, noisy, and highly susceptible to interference. This results in low stability and accuracy in detecting targets in three-dimensional space, and is prone to errors in obstacle classification. Furthermore, the detection process is prone to missed detections and false detections.

[0004] The technical problem solved by the present invention is how to improve the stability and accuracy of millimeter-wave radar detection and reduce the error in obstacle detection. Summary of the Invention

[0005] The present invention provides an obstacle detection method, device, and electronic equipment to address the defect of insufficient obstacle detection accuracy in the prior art, thereby improving the stability and accuracy of millimeter-wave radar detection and reducing the error in obstacle detection.

[0006] The present invention provides an obstacle detection method, comprising:

[0007] Collect radar point cloud data;

[0008] extracting feature information of each radar point from the radar point cloud data;

[0009] Inputting the characteristic information of each radar point into the obstacle detection model and outputting the obstacle detection result;

[0010] The obstacle detection model is trained using radar point cloud data samples corresponding to at least one obstacle with a label.

[0011] According to an obstacle detection method provided by the present invention, the characteristic information includes at least one of the following:

[0012] The radar point cloud includes position information, radar cross-section, pitch angle, heading angle, point cloud density, speed, and signal-to-noise ratio.

[0013] According to an obstacle detection method provided by the present invention, before inputting the characteristic information of each radar point into the obstacle detection model and outputting the obstacle detection result, the method further includes:

[0014] The semantic segmentation model is run using the extracted feature information of each radar point to output the semantic segmentation results of each radar point belonging to the obstacle category and background;

[0015] Perform DBSCAN clustering on the radar point cloud belonging to the obstacle class and output the target radar point cloud corresponding to the same type of obstacles;

[0016] Inputting the characteristic information of each radar point into the obstacle detection model and outputting the obstacle detection result includes:

[0017] The feature information of each radar point in the target radar point cloud is input into the obstacle detection model, and the obstacle detection result is output.

[0018] According to an obstacle detection method provided by the present invention, DBSCAN clustering is performed on radar point clouds belonging to obstacle classes, and target radar point clouds corresponding to obstacles of the same class are output, including:

[0019] Perform DBSCAN clustering on the radar point cloud belonging to the obstacle class and output the first bounding box of the target radar point cloud that surrounds the same type of obstacles;

[0020] Inputting the feature information of each radar point in the target radar point cloud into the obstacle detection model and outputting the obstacle detection results includes:

[0021] Inputting feature information of each radar point within the first bounding box into the obstacle detection model, and outputting a second bounding box;

[0022] The first bounding box and the second bounding box are added together to obtain the shape of the obstacle.

[0023] According to an obstacle detection method provided by the present invention, the semantic segmentation model and the obstacle detection model both adopt a multi-layer perceptron.

[0024] According to an obstacle detection method provided by the present invention, before extracting the feature information of each radar point from the radar point cloud data, the method further includes:

[0025] Fuse the point cloud data of the target number of frames collected;

[0026] Extract feature information of each radar point from the fused point cloud data.

[0027] According to an obstacle detection method provided by the present invention, before fusing the collected point cloud data of the target number of frames, the method further includes:

[0028] Based on a heuristic algorithm, the radar point cloud is preprocessed, including: removing noise.

[0029] The present invention also provides an obstacle detection device, comprising:

[0030] Acquisition module, used to collect radar point cloud data;

[0031] A feature extraction module is used to extract feature information of each radar point from the radar point cloud data;

[0032] An information processing module, configured to input the characteristic information of each radar point into an obstacle detection model and output an obstacle detection result;

[0033] The obstacle detection model is trained using radar point cloud data samples corresponding to at least one obstacle with a label.

[0034] An obstacle detection device according to the present invention further includes:

[0035] The semantic segmentation module is used to run the semantic segmentation model using the feature information of each radar point extracted, and output the semantic segmentation results of each radar point belonging to the obstacle category and background;

[0036] Clustering module, used to perform DBSCAN clustering on radar point clouds belonging to obstacle classes and output target radar point clouds corresponding to obstacles of the same class;

[0037] A calculation module is used to input the characteristic information of each radar point into the obstacle detection model and output the obstacle detection result, including:

[0038] The feature information of each radar point in the target radar point cloud is input into the obstacle detection model, and the obstacle detection result is output.

[0039] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-described obstacle detection methods are implemented.

[0040] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described obstacle detection methods.

[0041] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of any of the above obstacle detection methods are implemented.

[0042] The obstacle detection method and related equipment provided by the present invention extract features from each radar point in radar point cloud data to obtain information in multiple dimensions. This multi-dimensional feature information is then input into an obstacle detection model, which then detects and identifies obstacle types. By extracting features from single-point radar point cloud data, the present invention obtains more comprehensive radar point cloud data and more accurate feature information for each corresponding radar point, thereby reducing false detections and missed detections and improving the accuracy and stability of obstacle detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced one by one below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 This is one of the flow charts of the obstacle detection method provided by the present invention;

[0045] Figure 2 This is the second flow chart of the obstacle detection method provided by the present invention;

[0046] Figure 3 This is one of the structural diagrams of the obstacle detection device provided by the present invention;

[0047] Figure 4 This is the second structural diagram of the obstacle detection device provided by the present invention;

[0048] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0049] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0050] The following combination Figure 1-Figure 2 The obstacle detection method provided by the present invention is described.

[0051] The present invention provides an obstacle detection method, such as Figure 1 As shown, the following steps are included:

[0052] Step 110: Collect radar point cloud data;

[0053] The radar in this embodiment refers to a millimeter-wave radar, which can distinguish and identify very small targets and can identify multiple targets at the same time; it has the advantages of strong imaging capability, small size, good maneuverability and concealment.

[0054] In this embodiment, millimeter-wave radars are installed on the autonomous vehicle. Different numbers of radars can be set according to actual conditions, and they can be set at different positions on the autonomous vehicle to obtain radar point cloud data in different directions.

[0055] Step 120: extracting feature information of each radar point from the radar point cloud data;

[0056] Specifically, radar point clouds have multiple dimensions and complex distributions, making it difficult to uniformly process information from different dimensions. Therefore, it is necessary to perform single-point, local, and global feature extraction on the radar point cloud to obtain information from multiple dimensions. In this step, feature information from each radar point is extracted, i.e., single-point feature extraction is performed on the radar point cloud data.

[0057] Step 130: Inputting the characteristic information of each radar point into the obstacle detection model and outputting the obstacle detection result;

[0058] The obstacle detection model is trained using radar point cloud data samples corresponding to at least one obstacle with a label.

[0059] Specifically, the obstacle detection model detects and identifies objects represented by radar point cloud data. First, it determines whether the object is an obstacle or background. Obstacle types typically include vehicles and pedestrians, while background is typically a railing. Next, it further examines the obstacle's shape and size.

[0060] In this embodiment, the obstacle model is trained based on radar point cloud samples. Each training sample includes a target radar point cloud, and the target radar point cloud has been calibrated in the training sample. That is, this embodiment performs labeled training, which means that the target radar point cloud is known in each training sample. In this embodiment, the obstacle model is supervised by the training of the model based on multi-class cross entropy loss, ultimately achieving the function of semantic segmentation. Semantic segmentation is a fundamental task in computer vision. In semantic segmentation, we need to divide the visual input into different semantically interpretable categories. Semantic segmentation in this embodiment refers to segmenting the feature information of the radar point cloud into two categories: obstacles and background.

[0061] The obstacle detection method provided in this embodiment extracts features from radar point clouds to obtain information from multiple dimensions of the radar point cloud. This information is then fed into an obstacle detection model, which then detects and identifies obstacle types. By extracting features from single-point radar point cloud data, the present invention obtains more comprehensive radar point cloud data and more accurate feature information for each corresponding radar point. This reduces false detections and missed detections, improving the accuracy and stability of obstacle detection.

[0062] Optionally, the characteristic information includes at least one of the following:

[0063] The radar point cloud includes position information, radar cross-section, pitch angle, heading angle, point cloud density, speed, and signal-to-noise ratio.

[0064] Specifically, location information reflects the approximate orientation of the radar point cloud. Radar cross-section (RSC) refers to the ratio of the backscattered power per unit solid angle in the radar's direction of incidence to the power density of the target's cross-section. It characterizes the intensity of the echo generated by the radar wave. The pitch angle reflects the longitudinal angle of the obstacle relative to the autonomous vehicle. The heading angle reflects the lateral angle of the obstacle relative to the autonomous vehicle. Point cloud density reflects the density of radar echo reflections from the obstacle. Specifically, obstacles closer to the radar have denser point clouds, while those farther away have sparser point clouds. Point cloud velocity can be used as a basis for preliminary assessment of background and obstacle locations, and different dynamic obstacles have different velocities. The signal-to-noise ratio (SNR) refers to the ratio of signal to noise generated by the radar device. Here, the signal refers to the point cloud signal processed by the radar device, while the noise refers to the irregular, additional signals generated after passing through the radar device that are not present in the original signal.

[0065] The calculation formula of point cloud density is:

[0066]

[0067] Where ρ is the point cloud density, M is the number of radar point clouds in the selected area, and S is the area of ​​the radar point cloud in the selected area.

[0068] In another embodiment provided by the present invention, referring to Figure 2 , the obstacle detection method further includes the following steps:

[0069] Step 210: Run a semantic segmentation model using the extracted feature information of each radar point to output a semantic segmentation result of each radar point belonging to an obstacle category and background;

[0070] Step 220: Perform DBSCAN clustering on the radar point cloud belonging to the obstacle class, and output the target radar point cloud corresponding to the same type of obstacles;

[0071] Step 230: Inputting the characteristic information of each radar point into the obstacle detection model and outputting the obstacle detection result includes:

[0072] The feature information of each radar point in the target radar point cloud is input into the obstacle detection model, and the obstacle detection result is output.

[0073] Because radar point cloud detection is affected by the surrounding environment and the complexity of its multi-dimensional information, this embodiment first performs semantic segmentation on the feature information of the radar point cloud data, performs DBSCAN clustering on the radar point cloud belonging to the obstacle class, and then inputs the clustered radar point cloud into the obstacle detection model. This implementation further processes the radar point cloud of obstacles and background through semantic segmentation and DBSCAN clustering, thereby filtering out some radar point clouds that do not belong to obstacles before detection, thereby further improving detection accuracy.

[0074] Clustering algorithms are based on data similarity; patterns within a cluster are more similar than patterns in different clusters. In this embodiment, semantic segmentation and DBSCAN clustering are performed on the radar point cloud to filter out some background radar point clouds, narrowing the obstacle detection range and facilitating subsequent detection.

[0075] DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density clustering algorithm that determines clustering based on the density of sample distribution. By grouping closely connected samples into one cluster, a cluster classification is obtained. By grouping all groups of closely connected samples into different clusters, the final cluster classification results are obtained. In this example, each group of closely connected radar point clouds is grouped into one cluster, resulting in cluster classifications for multiple different groups of radar point clouds.

[0076] The DBSCAN clustering algorithm can be applied to both convex and non-convex sample sets. The significant advantages of the DBSCAN clustering algorithm are its fast clustering speed and its ability to effectively handle noise points and discover spatial clusters of arbitrary shapes.

[0077] The DBSCAN clustering algorithm is used to obtain the feature information of the bounding box of the radar point cloud corresponding to the obstacle. The feature information of the bounding box is input into the obstacle detection model to obtain more accurate detection results. In this embodiment, the obstacle detection model is supervised and trained using smooth L1 regression loss and binary cross entropy loss. The obstacle detection model outputs the center point of these bounding boxes and the extension compensation, and then adds the features of the bounding box before inputting the obstacle detection model to output the center point, length and width of the final bounding box, and outputs the confidence level.

[0078] This embodiment classifies the radar point clouds corresponding to each radar point, performs DBSCAN cluster analysis on the radar point clouds belonging to obstacles, and further extracts features. The features are then input into the obstacle detection model, which can further improve the detection accuracy and significantly reduce the false detection rate.

[0079] Optionally, performing DBSCAN clustering on the radar point clouds belonging to the obstacle class and outputting target radar point clouds corresponding to obstacles of the same class include:

[0080] Perform DBSCAN clustering on the radar point cloud belonging to the obstacle class and output the first bounding box of the target radar point cloud that surrounds the same type of obstacles;

[0081] Inputting the feature information of each radar point in the target radar point cloud into the obstacle detection model and outputting the obstacle detection results includes:

[0082] Inputting feature information of each radar point within the first bounding box into the obstacle detection model, and outputting a second bounding box;

[0083] The first bounding box and the second bounding box are added together to obtain the shape of the obstacle.

[0084] Since the bounding box is a virtual feature, the specific state of the bounding box cannot be displayed in the radar point cloud image. Therefore, it is impossible to directly extract features from the bounding box, and feature information processing and conversion are required.

[0085] Specifically, after obtaining the coordinates and other features of the bounding box, the feature extraction of the radar point cloud within the bounding box is performed, including the coordinate information, radar cross-sectional area, pitch angle, heading angle, point cloud density, and other information of the radar point cloud within the bounding box. The feature information of the radar point cloud within the bounding box is then used as the feature information of the bounding box and input into the obstacle detection model. The obstacle detection model outputs the detection result of the bounding box, which is added to the input bounding box to obtain the obstacle detection result.

[0086] In this embodiment, the semantic segmentation model and the obstacle detection model both use a multi-layer perceptron.

[0087] Optionally, before extracting the feature information of each radar point from the radar point cloud data, the method further includes:

[0088] Fuse the point cloud data of the target number of frames collected;

[0089] Extract feature information of each radar point from the fused point cloud data.

[0090] In this implementation, single-point radar point cloud extraction is performed frame by frame to obtain the radar point cloud corresponding to each frame. Simultaneously, motion compensation is performed on the radar point cloud corresponding to each frame using the vehicle's speed information. Multiple radar point clouds are then superimposed. This superimposed and fused radar point cloud increases the number of radar points, making feature extraction from the radar point cloud data more convenient.

[0091] Optionally, before fusing the collected point cloud data of the target number of frames, the method further includes:

[0092] Based on a heuristic algorithm, the radar point cloud is preprocessed, including: removing noise.

[0093] Specifically, a heuristic algorithm is an algorithm constructed based on intuition or experience, which provides a feasible solution for each instance of the combinatorial optimization problem to be solved at an acceptable cost (in terms of computing time and space). The degree of deviation between the feasible solution and the optimal solution is generally unpredictable. In this embodiment, the heuristic algorithm can be selected from: ant colony algorithm, simulated annealing method, neural network, etc.

[0094] Because radar point clouds are sparse, noisy, and highly interfering, in this embodiment, preprocessing conditions and thresholds are set to preprocess the radar point clouds, typically removing noise. Noise typically includes clutter and various active or passive interference within the system.

[0095] The specific method for removing noise is as follows: pre-process the point cloud collected by the radar, first convert the radar coordinate system to the vehicle coordinate system, and then set different radar cross-section (RCS) and signal-to-noise ratio (SNR) thresholds for radar points in different areas according to the radar's field of view (FOV) to perform adaptive noise removal.

[0096] This implementation preprocesses the radar point cloud to remove noise before feature extraction. The preprocessed radar point cloud can make subsequent feature extraction and detection results more accurate and convenient.

[0097] The obstacle detection device provided by the present invention is described below. The obstacle detection device described below and the obstacle detection method described above can be referenced to each other.

[0098] like Figure 3 As shown, the obstacle detection device provided by the present invention includes the following modules: an acquisition module 310 , a feature extraction module 320 and an information processing module 330 .

[0099] An acquisition module 310 is used to acquire radar point cloud data;

[0100] A feature extraction module 320 is used to extract feature information of each radar point from the radar point cloud data;

[0101] An information processing module 330 is configured to input the characteristic information of each radar point into an obstacle detection model and output an obstacle detection result;

[0102] The obstacle detection model is trained using radar point cloud data samples corresponding to at least one obstacle with a label.

[0103] The obstacle detection device provided in this embodiment extracts features from radar point clouds to obtain information from multiple dimensions of the radar point clouds. This information is then fed into an obstacle detection model, which then detects and identifies obstacle types. By extracting features from single-point radar point cloud data, the present invention obtains more comprehensive radar point cloud data and more accurate feature information for each corresponding radar point. This reduces false detections and missed detections, improving the accuracy and stability of obstacle detection.

[0104] In a preferred embodiment, if Figure 4 As shown, the obstacle detection device provided by this embodiment further includes: a semantic segmentation module 410 , a clustering module 420 and a calculation module 430 .

[0105] Semantic segmentation module 410, for running a semantic segmentation model using the extracted feature information of each radar point, and outputting a semantic segmentation result of each radar point belonging to an obstacle category and background;

[0106] Clustering module 420, configured to perform DBSCAN clustering on radar point clouds belonging to the obstacle class and output target radar point clouds corresponding to obstacles of the same class;

[0107] The computing module 430 is configured to input the characteristic information of each radar point into the obstacle detection model and output the obstacle detection result, including:

[0108] The feature information of each radar point in the target radar point cloud is input into the obstacle detection model, and the obstacle detection result is output.

[0109] The obstacle detection device provided in this embodiment can further improve detection accuracy and significantly reduce the false detection rate by semantically segmenting the radar point cloud corresponding to each radar point, performing DBSCAN cluster analysis and further feature extraction on the radar point cloud belonging to the obstacle, and then inputting the features into the obstacle detection model.

[0110] Figure 5 An example of a physical structure diagram of an electronic device is shown below. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the obstacle detection method, which includes:

[0111] Collect radar point cloud data;

[0112] extracting feature information of each radar point from the radar point cloud data;

[0113] Inputting the characteristic information of each radar point into the obstacle detection model and outputting the obstacle detection result;

[0114] The obstacle detection model is trained using radar point cloud data samples corresponding to at least one obstacle with a label.

[0115] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0116] In another aspect, the present invention further provides a computer program product, comprising a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the obstacle detection method provided by each of the above methods, which includes:

[0117] Collect radar point cloud data;

[0118] extracting feature information of each radar point from the radar point cloud data;

[0119] Inputting the characteristic information of each radar point into the obstacle detection model and outputting the obstacle detection result;

[0120] The obstacle detection model is trained using radar point cloud data samples corresponding to at least one obstacle with a label.

[0121] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the obstacle detection method provided by each of the above methods is implemented, and the method includes:

[0122] Collect radar point cloud data;

[0123] extracting feature information of each radar point from the radar point cloud data;

[0124] Inputting the characteristic information of each radar point into the obstacle detection model and outputting the obstacle detection result;

[0125] The obstacle detection model is trained using radar point cloud data samples corresponding to at least one obstacle with a label.

[0126] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0127] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An obstacle detection method, characterized in that: include: Collect radar point cloud data; extracting feature information of each radar point from the radar point cloud data; The semantic segmentation model is run using the extracted feature information of each radar point to output the semantic segmentation results of each radar point belonging to the obstacle category and background; Perform DBSCAN clustering on the radar point cloud belonging to the obstacle class and output the first bounding box of the target radar point cloud that surrounds the same type of obstacles; Inputting feature information of each radar point within the first bounding box into an obstacle detection model and outputting a second bounding box; Adding the first bounding box and the second bounding box to obtain the shape of the obstacle; The obstacle detection model is trained using radar point cloud data samples corresponding to at least one obstacle with a label.

2. The obstacle detection method according to claim 1, wherein: The characteristic information includes at least one of the following: The radar point cloud includes position information, radar cross-section, pitch angle, heading angle, point cloud density, speed, and signal-to-noise ratio.

3. The obstacle detection method according to claim 1, wherein: The semantic segmentation model and the obstacle detection model both adopt multi-layer perceptrons.

4. The obstacle detection method according to claim 1, wherein: Before extracting the feature information of each radar point from the radar point cloud data, the method further includes: Fuse the point cloud data of the target number of frames collected; Extract feature information of each radar point from the fused point cloud data.

5. The obstacle detection method according to claim 4, characterized in that: Before fusing the collected point cloud data of the target number of frames, the method further includes: Based on a heuristic algorithm, the radar point cloud is preprocessed, including: removing noise.

6. An obstacle detection device, characterized in that: include: Acquisition module, used to collect radar point cloud data; A feature extraction module is used to extract feature information of each radar point from the radar point cloud data; The semantic segmentation module is used to run the semantic segmentation model using the feature information of each radar point extracted, and output the semantic segmentation results of each radar point belonging to the obstacle category and background; The clustering module is used to perform DBSCAN clustering on the radar point cloud belonging to the obstacle class and output the first bounding box of the target radar point cloud surrounding the same type of obstacles; A calculation module is used to input the characteristic information of each radar point into the obstacle detection model and output the obstacle detection result, including: Inputting feature information of each radar point within the first bounding box into the obstacle detection model, and outputting a second bounding box; Adding the first bounding box and the second bounding box to obtain the shape of the obstacle; The obstacle detection model is trained using radar point cloud data samples corresponding to at least one obstacle with a label.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the obstacle detection method according to any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the obstacle detection method according to any one of claims 1 to 5 are implemented.

9. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the obstacle detection method according to any one of claims 1 to 5 are implemented.

Images