Obstacle detection method and device, electronic equipment and vehicle
By performing occupancy grid clustering and feature parameter decomposition on obstacles, the problem of inaccurate detection of bending shapes of large vehicles was solved, improving the accuracy and stability of obstacle detection for intelligent driving vehicles in complex scenarios and avoiding collisions with large vehicles.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-09
- Publication Date
- 2026-04-14
AI Technical Summary
Existing technology cannot accurately detect obstacles with curved shapes, especially large vehicles, which makes it easy for intelligent driving vehicles to have side scrapes or collisions in the "inner wheel difference" area when driving alongside or preparing to overtake large vehicles.
By acquiring obstacle information and performing occupancy grid clustering, characteristic parameters of large vehicles, such as bending points and longest axes, are identified. Obstacles are split into first and second parts and fused to obtain trackable obstacles, ensuring the accuracy and stability of detection.
It achieves accurate detection of unknown types of obstacles, improves the accuracy and stability of target detection in complex scenarios, and can effectively avoid collisions, especially in scenarios where large vehicles are turning.
Smart Images
Figure CN121849159A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of autonomous driving technology, and in particular to an obstacle detection method, device, electronic device, and vehicle. Background Technology
[0002] Large vehicles, due to their length and wheelbase, exhibit a unique "bending" shape when passing through intersections or turning. If the shape and space occupied by a bending vehicle cannot be accurately detected, autonomous vehicles are prone to side scrapes or collisions in the inner wheel difference area. Especially when autonomous vehicles are driving alongside large vehicles or preparing to overtake, accurate perception of the bending shape is crucial for making the right decisions.
[0003] In related technologies, deep learning networks, LiDAR, or fusion methods are typically used to detect unknown types of obstacles. The target object is modeled as a single, axis-aligned, or directional rectangle. However, a single rectangle is difficult to completely enclose a vehicle that is bent in an "L" or "V" shape. Therefore, it is impossible to accurately detect extremely long obstacles, which affects the safety of vehicle driving. Summary of the Invention
[0004] This application provides an obstacle detection method, apparatus, electronic device, and vehicle, aiming to solve the problem of inaccurate detection of unknown types of obstacles.
[0005] To address the aforementioned issues, this application discloses an obstacle detection method, comprising: acquiring obstacle information around a vehicle and corresponding occupancy grid clusters of objects; determining whether the occupancy grid clusters are large vehicles; if the occupancy grid clusters are large vehicles and the vehicle is in a turning scenario, acquiring feature parameters of the occupancy grid clusters, the feature parameters including: the turning point of the occupancy grid clusters and the longest axis of the occupancy grid clusters; splitting the occupancy grid clusters according to the feature parameters to obtain a first part and a second part; and fusing the first part and the second part to obtain the tracked obstacle.
[0006] Based on the above technical means, this application splits the occupied grid cluster objects by using the feature parameters of the occupied grid cluster objects to obtain a first part and a second part; the first part and the second part are fused to obtain the tracked obstacle, thereby enabling accurate detection of unknown types of obstacles to avoid obstacles, significantly improving the accuracy and stability of target detection in complex scenes. At the same time, by determining whether the occupied grid cluster objects are large vehicles, the turning scenes of large vehicles can be detected.
[0007] Optionally, the occupation grid cluster object is determined to be in a turning scene by: identifying the longest axis of the occupation grid cluster object and identifying the initial turning point of the occupation grid cluster object; calculating the comprehensive confidence of the initial candidate turning points; determining the core candidate points of the occupation grid cluster object according to the ranking of the comprehensive confidence; processing the core candidate points according to the set rules to obtain the turning point of the occupation grid cluster object; if the distance from the turning point to the longest axis of the occupation grid cluster object is greater than the set threshold, then the occupation grid cluster object is determined to be in a turning scene.
[0008] Based on the aforementioned technical means, it is possible to detect turning scenarios of large vehicles.
[0009] Optionally, processing the core candidate points according to a set rule to obtain the inflection point of the occupied grid cluster includes: determining whether the overall confidence score of each core candidate point is greater than a confidence score threshold; if the overall confidence score of the core candidate points is greater than the confidence score threshold, then the core candidate points with overall confidence scores greater than the confidence score threshold are weighted and averaged according to their overall confidence scores to obtain the inflection point; if the overall confidence scores of all core candidate points are less than the confidence score threshold, then the grid farthest from the longest axis of the occupied grid cluster is taken as the inflection point.
[0010] Based on the above technical means, it can be ensured that the disassembly results conform to the actual structure of large vehicles, thereby improving the accuracy of detection.
[0011] Optionally, splitting the occupied grid cluster object according to the feature parameters to obtain a first part and a second part includes: traversing all grids in the occupied grid cluster object; calculating a first vector from the bend point to each of the occupied grids; calculating the dot product of the direction vector of the longest axis and the first vector, where the longest axis is the line connecting the two grids farthest apart in the occupied grid cluster object; if the dot product is greater than zero, then determining that the current grid belongs to the first direction of the longest axis, and taking all grids belonging to one side of the first direction as the first part; if the dot product is less than zero, then determining that the current grid belongs to the second direction of the longest axis, and taking all grids belonging to one side of the second direction as the second part.
[0012] Based on the aforementioned technical means, it is possible to accurately classify unknown types of obstacles.
[0013] Optionally, fusing the first part and the second part to obtain the tracked obstacle includes: matching the obstacle information with the occupied grid cluster objects to obtain an obstacle information matching the occupied grid cluster objects; calculating a first overlap between the obstacle information and the first part; if the first overlap is greater than a set overlap threshold, then determining that the first part and the obstacle information correspond to the same obstacle; determining a first detection box corresponding to the first part based on the attribute information of the obstacle information and the attribute information of the first part; calculating a second overlap between the obstacle information and the second part; if the second overlap is less than a set overlap threshold, then using the attribute information of the second part as a second detection box corresponding to the second part; and obtaining the tracked obstacle based on the first detection box and the second detection box.
[0014] Based on the above-mentioned technical means, accurate detection, complete tracking, and precise size prediction of tracking obstacles can be achieved.
[0015] Optionally, fusing the first part and the second part to obtain tracked obstacles includes: matching the obstacle information with the occupied grid cluster objects to obtain two obstacle information sets matched with the occupied grid cluster objects, the two obstacle information sets being the first obstacle information and the second obstacle information; calculating the third overlap between the first obstacle information and the first part; if the third overlap is greater than a set overlap threshold, then determining that the first obstacle information and the first part correspond to the same obstacle; determining the third detection box corresponding to the first part based on the attribute information of the first obstacle information and the attribute information of the first part; if the third overlap is less than a set threshold, calculating the fourth overlap between the first obstacle and the second part; if the fourth overlap is greater than a set overlap threshold, then determining that the first obstacle information and the second part correspond to the same obstacle; and determining the third detection box corresponding to the first part based on the attribute information of the first obstacle information. The system determines a fourth detection box corresponding to the second part based on the attribute information of the second obstacle and the attribute information of the second part; calculates a fifth overlap between the second obstacle information and the first part; if the fifth overlap is greater than a set overlap threshold, it determines that the second obstacle information and the first part correspond to the same obstacle; determines a fifth detection box corresponding to the first part based on the attribute information of the second obstacle information and the attribute information of the first part; if the fifth overlap is less than a set threshold, it calculates a sixth overlap between the second obstacle and the second part; if the sixth overlap is greater than a set overlap threshold, it determines that the second obstacle information and the second part correspond to the same obstacle; determines a sixth detection box corresponding to the second part based on the attribute information of the second obstacle information and the attribute information of the second part; and obtains the tracking obstacle based on the third detection box and the sixth detection box, or based on the fourth detection box and the fifth detection box.
[0016] Based on the above-mentioned technical means, accurate detection, complete tracking, and precise size prediction of tracking obstacles can be achieved.
[0017] Optionally, fusing the first part and the second part to obtain the tracked obstacle includes: matching the obstacle information with the occupied grid cluster objects to obtain no obstacle information matching the occupied grid cluster objects; calculating the seventh overlap between the first part and historical tracked obstacles in the tracked obstacle list; if the seventh overlap is less than a set threshold, creating a third obstacle information for the first part according to the attribute information of the first part; calculating the eighth overlap between the second part and historical tracked obstacles in the tracked obstacle list; if the eighth overlap is less than a set threshold, creating a fourth obstacle information for the second part according to the attribute information of the second part; and obtaining the tracked obstacle based on the third obstacle information and the fourth obstacle information.
[0018] Based on the above-mentioned technical means, accurate detection, complete tracking, and precise size prediction of tracking obstacles can be achieved.
[0019] To address the aforementioned problems, this application also discloses an obstacle detection device, comprising: The acquisition module is used to acquire obstacle information around the vehicle and the occupancy grid clustering objects corresponding to the obstacle information; The first judgment module is used to determine whether the object occupying the grid cluster is a large vehicle; The parameter module is used to obtain the feature parameters of the occupying grid cluster if the occupying grid cluster is a large vehicle and is in a turning scene. The feature parameters include: the turning point of the occupying grid cluster and the longest axis of the occupying grid cluster. The splitting module is used to split the occupied grid clustered object according to the feature parameters to obtain a first part and a second part; The fusion module is used to fuse the first part and the second part to obtain the tracked obstacle.
[0020] To address the aforementioned problems, this application also discloses an electronic device, including a processor and a memory, wherein... Memory, used to store computer programs; The processor, when executing a computer program stored in memory, implements the obstacle detection method described above.
[0021] To address the aforementioned issues, this application also discloses a vehicle containing a computer program that, when executed by a processor, implements the aforementioned obstacle detection method.
[0022] Of course, any product implementing this application does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description
[0023] Figure 1 This is a flowchart of an obstacle detection method provided in an embodiment of this application; Figure 2 This is a flowchart of an obstacle detection method provided in an embodiment of this application; Figure 3 This is a schematic diagram of the splitting of the grid clustered objects occupied by this application; Figure 4 This is a flowchart of an obstacle detection method provided in an embodiment of this application; Figure 5-a This is a schematic diagram illustrating the matching of an obstacle with the occupied grid clustered object according to an embodiment of this application; Figure 5-b This is a schematic diagram of the optimized obstacle tracking according to an embodiment of this application; Figure 6 This is a flowchart of an obstacle detection method provided in an embodiment of this application; Figure 7-a This is a schematic diagram of two tracking obstacles and the occupied grid clustered objects according to an embodiment of this application; Figure 7-b This is a schematic diagram of the optimized obstacle tracking according to an embodiment of this application; Figure 8 This is a flowchart of an obstacle detection method provided in an embodiment of this application; Figure 9-a This is a schematic diagram illustrating the absence of tracking obstacles matching the clustered objects in the occupied grid in this embodiment of the application; Figure 9-b This is a schematic diagram of the optimized obstacle tracking according to an embodiment of this application; Figure 10 This is a structural diagram of an obstacle detection device provided in an embodiment of this application; Figure 11 This is a structural diagram of the electronic device provided in the embodiments of this application. Detailed Implementation
[0024] To make the technical problems, technical solutions, and beneficial effects solved by this application clearer, the following detailed description is provided in conjunction with embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0025] Example 1 refer to Figure 1 The diagram shows a flowchart of an obstacle detection method provided in an embodiment of this application. The method includes the following steps: Step 101: Obtain information on obstacles around the vehicle and the corresponding occupied grid clustered objects. In practical applications, sensors such as LiDAR and millimeter-wave radar can be used to collect image information around the vehicle. This image information is then input into a neural network model to obtain the obstacle's position information (center point), size, obstacle detection box, and heading angle in the vehicle coordinate system. The size includes the obstacle's length, width, and height. The obstacle's position information in the vehicle coordinate system, its size, and its category are then stored in an obstacle list. Obstacles include, but are not limited to, vehicles, pedestrians, traffic signs, flower beds, etc.
[0026] The obstacle list is processed using a target tracking algorithm to obtain tracked obstacles. Compared to regular obstacles, tracked obstacles have additional information such as tracking ID, speed, and acceleration. Then, the position information of the tracked obstacles in the vehicle coordinate system, the size, category, tracking ID, speed, and acceleration information are stored in the tracked obstacle list.
[0027] The obstacle information is processed using the Occupancy Grid Map (OCC) segmentation model to obtain the occupancy grid information corresponding to the obstacle information. This occupancy grid information includes information such as the size and type of each grid.
[0028] The occupied grid information is clustered according to the occupied grid information belonging to the same category within eight neighborhoods to obtain a cluster object. This cluster object corresponds to the same physical entity and is marked with a cluster identifier.
[0029] For example, a turning semi-trailer is represented as a continuous occupied grid in OCC. Through eight-neighbor clustering, these grids will be grouped into a single cluster, which corresponds to the semi-trailer. The adjacent flower bed is another continuous occupied grid, which will be grouped into a separate cluster, which corresponds to the flower bed.
[0030] Calculate the geometric features of each cluster object, which include: center point, size, and heading angle.
[0031] The clustered objects are classified according to the attributes of the original grid corresponding to each clustered object. The categories include: static, dynamic, traversable area, etc. Among them, dynamic represents moving targets such as vehicles and pedestrians, static represents immovable targets such as traffic warning objects and flower beds, and traversable area represents drivable areas such as road surface.
[0032] The above operations yield multiple occupied grid cluster objects. Each occupied grid cluster object includes: cluster ID, center point, size, heading angle, and category.
[0033] For example: Occupying grid cluster object 1: ID=1, center point=(x1,y1), length=15m (extra long), heading angle=30°, category=dynamic (corresponding to a turning semi-trailer).
[0034] Occupying grid cluster object 1: ID=2, center point=(x2,y2), length=2m, heading angle=0°, category=static (corresponding to flower bed).
[0035] Step 102: Determine whether the object occupying the grid cluster is a large vehicle.
[0036] Determine if the size of the object occupying the grid cluster is greater than the set length threshold. If the size of the object occupying the grid cluster is greater than the length threshold, and the number of grids contained in the object occupying the grid cluster is also greater than the number threshold, then the object occupying the grid cluster can be considered a large vehicle. A large vehicle can be understood as a large vehicle, a long vehicle, or an extra-long vehicle, etc. Generally, the length of a large vehicle is greater than 22 meters, and the width of a large vehicle is less than 3 meters. For example, large vehicles are trailers, large buses, cement concrete transport vehicles, etc.
[0037] Step 103: If the occupied grid clustered object is a large vehicle and is in a turning scene, obtain the feature parameters of the occupied grid clustered object.
[0038] The feature parameters include the bend points of the occupied grid cluster and the longest axis of the occupied grid cluster, wherein the longest axis is the line connecting the two furthest grids in the occupied grid cluster.
[0039] For example, in the grid cluster of objects corresponding to obstacles, find the two grids that are furthest apart, namely P1 and P2, and take the line connecting P1 and P2 as the longest axis. This longest axis is the main direction axis.
[0040] Step 104: Split the clustered objects occupying the grid according to the feature parameters to obtain the first part and the second part.
[0041] Specifically, based on the bending point and longest axis of the occupying grid cluster object, the occupying grid cluster object is split into two parts, namely the first part and the second part.
[0042] Step 105: Perform a fusion process on the first part and the second part to obtain the tracked obstacle.
[0043] In this embodiment, the occupies grid cluster objects are split into a first part and a second part by using the feature parameters of the occupies grid cluster objects. The first part and the second part are then fused to obtain the tracked obstacle, thereby enabling accurate detection and avoidance of unknown types of obstacles. This significantly improves the accuracy and stability of target detection in complex scenes. At the same time, by determining whether the occupies grid cluster objects are large vehicles, the turning scenes of large vehicles can be detected.
[0044] Example 2 refer to Figure 2 The diagram shows a flowchart of an obstacle detection method provided in an embodiment of this application. The method includes the following steps: Step 201: Obtain information on obstacles around the vehicle and the corresponding occupancy grid clustered objects. Step 202: Determine whether the object occupying the grid cluster is a large vehicle. If the object occupying the grid cluster is a large vehicle, then proceed to step 203.
[0045] Determine whether the object occupying the grid cluster is a large vehicle. If the object occupying the grid cluster is a large vehicle, then determine whether the object occupies the grid cluster in a turning scene. If the object occupies the grid cluster is not a large vehicle, then the process ends.
[0046] Step 203: Determine whether the occupied grid clustered object is in a turning scene. If the occupied grid clustered object is in a turning scene, then proceed to step 204.
[0047] Step 204: Obtain the feature parameters of the occupied grid clustered objects.
[0048] Step 205: Split the occupied grid clustered object according to the feature parameters to obtain the first part and the second part.
[0049] Step 206: Perform a fusion process on the first part and the second part to obtain the tracked obstacle.
[0050] As one implementation method, step 203 includes the following sub-steps: Sub-step 2031: Identify the longest axis of the occupied grid clustered object.
[0051] Sub-step 2032: Identify the initial candidate bend points of the occupied grid clustered objects.
[0052] Sub-step 2033: Calculate the overall confidence level of the initial candidate bending points.
[0053] Sub-step 2034: Determine the core candidate points of the occupied grid cluster objects based on the ranking of the comprehensive confidence scores.
[0054] Sub-step 2035: Process the core candidate points according to the set rules to obtain the bending points.
[0055] Sub-step 2036: If the distance from the bending point to the longest axis of the occupied grid cluster object is greater than a set threshold, then the occupied grid cluster object is determined to be in a turning scene.
[0056] Sub-step 2032 identifies the initial candidate bend points of objects occupying the grid cluster as follows: In this embodiment, in the occupied grid cluster, the coordinates of the center points of all occupied grids are selected, the distance between any two grid center points is calculated, the two grids with the farthest distance (P1, P2) are found, and the line connecting P1 and P2 is determined as the longest axis of the occupied grid cluster, i.e., the main direction axis. P1 is defined as the grid with the smaller x-axis coordinate value, and P2 is defined as the grid with the larger x-axis coordinate value.
[0057] The origin is the midpoint of the longest axis, where the coordinates of the midpoint are (P1 + P2) / 2.
[0058] Calculate the axial direction vector and the normal vector. The axial direction vector is P1 pointing to P2, and the normal vector is the perpendicular vector of the axial direction vector at its midpoint. Then calculate the distance (projected distance along the normal direction) from the center point of each grid cell of the grid clustered object to the midpoint of the longest axis. Sort the distances from largest to smallest and select grid cells of a preset proportion as initial candidate bending points, for example, the preset proportion is 10%.
[0059] Sub-step 2033: The method for calculating the overall confidence level of the initial candidate bending points is as follows: To calculate the overall confidence level of the initial candidate bending points, in practical applications, the initial candidate bending points can be weighted by multi-dimensional scoring to obtain the overall confidence level.
[0060] The multi-dimensional scoring includes: location score, distance score, and consistency score.
[0061] The position score is determined based on the relative position of the initial candidate bend point and the midpoint of the longest axis.
[0062] The distance score is determined based on the ratio of the normal direction projection distance of the initial candidate bending point to the maximum normal direction projection distance.
[0063] The consistency score is determined based on the degree of continuity of the distribution of the grids surrounding the initial candidate bend points.
[0064] The initial candidate bend points are weighted according to their position score, distance score, and consistency score to obtain a comprehensive confidence score. Then, the comprehensive confidence scores are sorted in descending order to determine the core candidate points of the grid clustered objects. Specifically, the top 3 or top 5 initial candidate bend points with the highest comprehensive confidence scores can be selected as core candidate points.
[0065] Sub-step 2035 processes the core candidate points according to the set rules to obtain the bending points of the objects occupying the grid cluster as follows: Determine whether the overall confidence level of each core candidate point is greater than the confidence level threshold; if the overall confidence level of the core candidate point is greater than the confidence level threshold, then take a weighted average of the core candidate points with an overall confidence level greater than the confidence level threshold to obtain the bending point of the target obstacle, and use the bending point to ensure that the splitting result conforms to the actual vehicle structure of the large vehicle.
[0066] If the overall confidence score of all the core candidate points is less than the confidence score threshold, then the grid cell that is furthest from the longest axis of the occupied grid cluster object is taken as the bend point.
[0067] For example: Select the top 3 core candidate points with the highest overall confidence. If the overall confidence of these 3 core candidate points is >0.5 (confidence threshold), calculate the weighted average of these 3 core candidate points according to their overall confidence to obtain the final bending point.
[0068] If the combined confidence of the first 3 core candidate points is ≤0.5 (no suitable core candidate points), then the grid cell that occupies the longest axis of the grid clustered object is directly taken as the bending point (the point that is most far from the main axis is the bending point by default).
[0069] Step 205 includes the following sub-steps: Sub-step 2051: Traverse all occupied graticles in the occupied graticle cluster object.
[0070] Sub-step 2052: Calculate the first vector from the bend point to each of the all occupied grids.
[0071] Sub-step 2053: Calculate the dot product of the direction vector of the longest axis and the first vector, where the longest axis is the line connecting the two furthest grid cells in the occupied grid cluster object.
[0072] Sub-step 2054: If the dot product result is greater than zero, then determine that the current grid belongs to the first direction of the feature parameter, and take all grids belonging to one side of the first direction as the first part.
[0073] Sub-step 2055: If the dot product result is less than zero, then determine that the current grid belongs to the second direction of the feature parameter, and take all grids belonging to one side of the second direction as the second part.
[0074] Specifically, iterate through all occupied graticles contained in the occupied graticle cluster object, and calculate the vector l from the bend point to the center point of each graticle. i , where i represents the grid number, i=1,2,3,…,n. Then, by calculating the vector dot product of li and the longest axis, the position of the current grid relative to the bend point is determined based on the dot product result: if the dot product result is greater than 0, it means that the current grid is on the positive direction (first direction) of the longest axis, and all grids belonging to the first direction are taken as the first part; if the dot product result is less than 0, it means that the current grid is on the negative direction (second direction) of the longest axis, and all grids belonging to the second direction are taken as the second part. The first part constitutes set F, and the second part constitutes set R.
[0075] exist Figure 3 In the diagram, green grids represent dynamic OCC clustered objects, red grids represent static OCC clustered objects, blue grids represent passable areas, and black grids represent unknowns. The system determines whether the size of a dynamically occupied grid clustered object exceeds a set length threshold. If the size of the dynamically occupied grid clustered object exceeds the length threshold, and the number of grids it contains also exceeds the number threshold, then the dynamically occupied grid clustered object can be considered a large vehicle. Next, it determines whether the dynamically occupied grid clustered object is in a turning scenario. If the dynamically occupied grid clustered object is in a turning scenario, it is split into a first part 301 and a second part 302 based on its turning point and longest axis. The first part 301 and the second part 302 are then fused to obtain the tracked obstacle.
[0076] Example 3 refer to Figure 4 The flowchart of an obstacle detection method provided in an embodiment of this application is shown, including the following steps: Step 401: Obtain information on obstacles around the vehicle and the corresponding occupancy grid clustered objects.
[0077] Step 402: Determine whether the object occupying the grid cluster is a large vehicle.
[0078] Step 403: If the object occupying the grid cluster is a large vehicle and is in a turning scene, obtain the feature parameters of the object occupying the grid cluster.
[0079] Step 404: Split the occupied grid clustered objects according to the feature parameters to obtain the first part and the second part.
[0080] Step 405: Match the obstacle information with the occupied grid cluster objects to obtain an obstacle information matching the occupied grid cluster objects.
[0081] In practical applications, obstacle information is matched with multiple occupied grid clusters. If only one obstacle matches an occupied grid cluster, it's necessary to determine whether the obstacle matches the first or second part of the occupied grid cluster. If it matches the first part, it means the obstacle belongs to the same obstacle; otherwise, it matches a different obstacle. Similarly, if the obstacle matches the second part, it means the obstacle belongs to the same obstacle; otherwise, it matches a different obstacle.
[0082] Step 406: Calculate the first overlap between the obstacle information and the first part. If the first overlap is greater than a set overlap threshold, then determine that the first part and the obstacle information correspond to the same obstacle.
[0083] Step 407: Determine the first detection box corresponding to the first part based on the attribute information of the obstacle information and the attribute information of the first part.
[0084] If the first degree of overlap is less than or equal to the set overlap threshold, it means that the first part does not correspond to the same obstacle as the obstacle information, and the overlap between the obstacle and the second part is then determined.
[0085] The obstacle information attributes include: the center point coordinates of the obstacle detection box, the size of the obstacle detection box, the heading angle of the obstacle detection box, and the category of the obstacle detection box.
[0086] The attributes in the first part include: the center point, length, and width of the raster cluster, the heading angle of the longest axis, and the category.
[0087] In practical applications, determining the first detection box corresponding to the first part based on the attribute information of the obstacle and the attribute information of the first part includes: calculating the confidence score of the first part based on the attribute information of the first part; calculating the confidence score of the obstacle based on the attribute information of the obstacle; and calculating the fusion weight based on the confidence scores of the first part and the obstacle, where the fusion weight = confidence score of the first part / (confidence score of the first part + confidence score of the obstacle).
[0088] Based on the fusion weight, the attribute information of the first part and the attribute information of the obstacle are combined to determine the attribute information of the first detection box corresponding to the first part. The attribute information of the first detection box = (1-fusion weight) * obstacle attribute information + fusion weight * attribute information of the first part. Each detection box will have a corresponding obstacle, and each detection box can be understood as an obstacle detection box.
[0089] For example, if the attribute information of the obstacle and the attribute information of the first part are length information, then determining the first detection box corresponding to the first part includes: calculating the confidence score of the first part based on the reasonableness of its size and aspect ratio; calculating the confidence score of the obstacle based on its length and motion consistency; and calculating the fusion weight based on the confidence scores of the first part and the obstacle, where fusion weight = confidence score of the first part / (confidence score of the first part + confidence score of the obstacle).
[0090] Based on the fusion weight, the attributes of the first detection box corresponding to the first part are determined by combining the attribute information of the first part and the attribute information of the obstacle. The attribute information of the first detection box is (1-fusion weight)*the position of the obstacle + fusion weight*the position of the first part.
[0091] The attributes of obstacles also include the heading angle or category of the obstacle detection box, which can be determined in the same way as described above, and will not be explained in detail here.
[0092] Step 408: Calculate the second overlap between the obstacle information and the second part. If the second overlap is less than the set overlap threshold, then use the attribute information of the second part as the second detection box corresponding to the second part.
[0093] In practical applications, if the second overlap is less than the set overlap threshold, it means that the obstacle information corresponds to a different obstacle than the second part. In this case, the attribute information of the second part can be directly used as the attribute information of the second detection box corresponding to the second part. The second detection box corresponds to an obstacle, and each detection box can be understood as an obstacle detection box.
[0094] Step 409: Obtain the tracked obstacle based on the first detection box and the second detection box.
[0095] For example: Matching obstacle 501 with multiple occupied grid clusters, it was found that only one obstacle matched an object in the occupied grid cluster. Figure 5-a As shown, Figure 5-a This is a schematic diagram of an obstacle and an object clustered in an occupied grid according to an embodiment of this application. After processing using the methods in steps 406-409, a first detection box and a second detection box are obtained, as shown below. Figure 5-bAs shown, 502 represents the first detection box and 503 represents the second detection box. The first and second detection boxes are used as tracking obstacles.
[0096] In this embodiment, the accuracy of obstacle detection is improved by optimizing the tracking of obstacles.
[0097] Example 4 refer to Figure 6 The flowchart of an obstacle detection method provided in an embodiment of this application is shown, including the following steps: Step 601: Obtain obstacle information around the vehicle and the occupancy grid clustering objects corresponding to the obstacle information.
[0098] Step 602: Determine whether the object occupying the grid cluster is a large vehicle.
[0099] Step 603: If the object occupying the grid cluster is a large vehicle and is in a turning scene, obtain the feature parameters of the object occupying the grid cluster.
[0100] Step 604: Split the occupied grid clustered objects according to the feature parameters to obtain a first part and a second part.
[0101] Step 605: Match the obstacle information with the occupied grid cluster objects to obtain two obstacle information matching the occupied grid cluster objects.
[0102] The two obstacle information are respectively the first obstacle information and the second obstacle information.
[0103] In practical applications, obstacle information is matched with multiple occupied grid clusters. If two obstacle information matches an occupied grid cluster, it's necessary to determine whether these two obstacle information belong to the first or second part of the occupied grid cluster. If the obstacle information matches the first part, it means they correspond to the same obstacle; if they don't, they correspond to different obstacles. Similarly, if the obstacle information matches the second part, it means they correspond to the same obstacle; if they don't, they correspond to different obstacles.
[0104] Step 606: Calculate the overlap between the two obstacle information and the first and second parts, respectively.
[0105] Step 606 includes the following sub-steps: Sub-step 6061: Calculate the third overlap between the first obstacle information and the first part. If the third overlap is greater than a set overlap threshold, then determine that the first obstacle information and the first part correspond to the same obstacle.
[0106] Sub-step 6062: Determine the third detection box corresponding to the first part based on the attribute information of the first obstacle information and the attribute information of the first part.
[0107] The attribute information of the first obstacle includes: the center point coordinates of the obstacle detection box, the size of the obstacle detection box, the heading angle of the obstacle detection box, and the category of the obstacle detection box.
[0108] The attributes in the first part include: the center point, length, and width of the raster cluster, the heading angle of the longest axis, and the category.
[0109] In practical applications, determining the third detection box corresponding to the first part, based on the attribute information of the obstacle and the attribute information of the first part, includes: calculating the confidence score of the first part based on its attribute information; calculating the confidence score of the obstacle based on its attribute information; and calculating the fusion weight based on the confidence scores of the first part and the obstacle, where fusion weight = confidence score of the first part / (confidence score of the first part + confidence score of the obstacle).
[0110] Based on the fusion weight, the attribute information of the first part and the attribute information of the obstacle are combined to determine the attribute information of the third detection box corresponding to the first part. The attribute information of the third detection box = (1 - fusion weight) * obstacle attribute information + fusion weight * attribute information of the first part.
[0111] For example, if the attribute information of the first obstacle and the attribute information of the first part are length information, then determining the third detection box corresponding to the first part includes: calculating the confidence score of the first part based on the reasonableness of its size and aspect ratio; calculating the confidence score of the first obstacle based on its length and motion consistency; and calculating the fusion weight based on the confidence scores of the first part and the first obstacle, where fusion weight = confidence score of the first part / (confidence score of the first part + confidence score of the first obstacle).
[0112] Based on the fusion weight, and combining the attribute information of the first part and the attribute information of the first obstacle, the attributes of the third detection box corresponding to the first part are determined. The attribute information of the third detection box is calculated as (1 - fusion weight) * position of the first obstacle + fusion weight * position of the first part. Each third detection box corresponds to one obstacle, and each detection box can be understood as an obstacle detection box. The attributes of the first obstacle also include the heading angle or category of the obstacle detection box, which can be determined in the same way as described above.
[0113] Sub-step 6063: If the third overlap is less than a set threshold, then calculate the fourth overlap between the first obstacle and the second part; if the fourth overlap is greater than the set overlap threshold, then determine that the first obstacle information and the second part correspond to the same obstacle.
[0114] Sub-step 6064: Determine the fourth detection box corresponding to the second part based on the attribute information of the first obstacle information and the attribute information of the second part.
[0115] In practical applications, determining the fourth detection box corresponding to the second part based on the attribute information of the first obstacle and the attribute information of the second part includes: calculating the confidence score of the second part based on the attribute information of the second part; calculating the confidence score of the first obstacle based on the attribute information of the first obstacle; and calculating the fusion weight based on the confidence scores of the second part and the first obstacle, where the fusion weight = confidence score of the second part / (confidence score of the second part + confidence score of the first obstacle).
[0116] Based on the fusion weight, the attribute information of the second part and the attribute information of the first obstacle are combined to determine the attribute information of the fourth detection box corresponding to the second part. The attribute information of the fourth detection box = (1 - fusion weight) * attribute information of the first obstacle + fusion weight * attribute information of the second part.
[0117] Sub-step 6065: Calculate the fifth overlap between the second obstacle information and the first part. If the fifth overlap is greater than a set overlap threshold, then determine that the second obstacle information and the first part correspond to the same obstacle.
[0118] Sub-step 6066: Determine the fifth detection box corresponding to the first part based on the attribute information of the second obstacle information and the attribute information of the first part.
[0119] In practical applications, determining the fourth detection box corresponding to the first part based on the attribute information of the second obstacle and the attribute information of the first part includes: calculating the confidence score of the first part based on the attribute information of the first part; calculating the confidence score of the second obstacle based on the attribute information of the second obstacle; and calculating the fusion weight based on the confidence scores of the first part and the second obstacle, where the fusion weight = confidence score of the first part / (confidence score of the first part + confidence score of the second obstacle).
[0120] Based on the fusion weight, the attribute information of the first part and the attribute information of the second obstacle are combined to determine the attribute information of the fifth detection box corresponding to the first part. The attribute information of the fifth detection box = (1 - fusion weight) * attribute information of the second obstacle + fusion weight * attribute information of the first part.
[0121] Sub-step 6067: If the fifth overlap is less than a set threshold, then calculate the sixth overlap between the second obstacle and the second part; if the sixth overlap is greater than the set overlap threshold, then determine that the second obstacle information and the second part correspond to the same obstacle.
[0122] Sub-step 6068: Determine the sixth detection box corresponding to the second part based on the attribute information of the second obstacle information and the attribute information of the second part.
[0123] The second obstacle information includes the following attributes: the center point coordinates of the obstacle detection box, the size of the obstacle detection box, the heading angle of the obstacle detection box, and the category of the obstacle detection box.
[0124] The attributes in the second part include: the center point, length, and width of the raster cluster, the heading angle of the longest axis, and the category.
[0125] In practical applications, determining the sixth detection box corresponding to the second part based on the attribute information of the second obstacle and the attribute information of the second part includes: calculating the confidence score of the second part based on the attribute information of the second part; calculating the confidence score of the second obstacle based on the attribute information of the second obstacle; and calculating the fusion weight based on the confidence scores of the second part and the second obstacle, where the fusion weight = confidence score of the second part / (confidence score of the second part + confidence score of the second obstacle).
[0126] Based on the fusion weight, the attribute information of the second part and the attribute information of the second obstacle are combined to determine the attribute information of the fifth detection box corresponding to the second part. The attribute information of the fifth detection box = (1 - fusion weight) * attribute information of the second obstacle + fusion weight * attribute information of the second part.
[0127] Sub-step 6069: Obtain the tracked obstacle based on the third detection box and the sixth detection box, or obtain the tracked obstacle based on the fourth detection box and the fifth detection box.
[0128] Matching obstacles with multiple occupied grid clusters revealed that only two obstacles matched objects in the occupied grid cluster, such as... Figure 7-a As shown, in Figure 7-a In this diagram, 701 represents the first obstacle information, and 702 represents the second obstacle information. After processing using the methods in steps 605-614, the third and sixth detection boxes are obtained as follows: Figure 7-b As shown, 703 represents the third detection box and 704 represents the sixth detection box. The third and sixth detection boxes are used as tracking obstacles.
[0129] In this embodiment, the accuracy of obstacle detection is improved by optimizing the tracking of obstacles.
[0130] Example 5 refer to Figure 8 The flowchart of an obstacle detection method provided in an embodiment of this application is shown, including the following steps: Step 801: Obtain obstacle information around the vehicle and the corresponding occupied grid cluster objects of the obstacle information.
[0131] Step 802: Determine whether the object occupying the grid cluster is a large vehicle.
[0132] Step 803: If the object occupying the grid cluster is a large vehicle and is in a turning scene, obtain the feature parameters of the object occupying the grid cluster.
[0133] Step 804: Split the occupied grid clustered object according to the feature parameters to obtain the first part and the second part.
[0134] Step 805: Match the obstacle information with the occupied grid cluster objects to obtain no obstacle information that matches the occupied grid cluster.
[0135] Step 806: Calculate the seventh overlap between the first part and the historical tracked obstacles in the tracked obstacle list.
[0136] Step 807: If the seventh overlap is less than a set threshold, then create a third obstacle information for the first part according to the attribute information of the first part.
[0137] Step 808: Calculate the eighth overlap between the second part and the historical tracked obstacles in the tracked obstacle list.
[0138] Step 809: If the eighth overlap is less than the set threshold, then create a fourth obstacle information for the second part according to the attribute information of the second part.
[0139] Step 810: Obtain the tracked obstacle based on the third obstacle information and the fourth obstacle information.
[0140] In practical applications, the seventh overlap degree between the first part and the first historical tracked obstacle in the obstacle tracking list is calculated. If the seventh overlap degree is greater than a set threshold, it indicates that the first historical tracked obstacle and the first part are the same obstacle. In this case, the seventh detection box corresponding to the first part is determined based on the attribute information of the first part and the attribute information of the first historical tracked obstacle. If the seventh overlap degree is less than the set threshold, a third obstacle information is created for the first part according to the attribute information of the first part.
[0141] In practical applications, determining the seventh detection box corresponding to the first part based on the attribute information of the first part and the attribute information of the first historical tracked obstacles includes: calculating the confidence score of the first part based on the attribute information of the first part; calculating the confidence score of the first historical tracked obstacles based on the attribute information of the first historical tracked obstacles; and calculating the fusion weight based on the confidence scores of the first part and the first historical tracked obstacles, where the fusion weight = confidence score of the first part / (confidence score of the first part + confidence score of the first historical tracked obstacles).
[0142] Based on the fusion weight, and combining the attribute information of the first part and the attribute information of the first historical tracked obstacle, the attribute information of the seventh detection box corresponding to the first part is determined. The attribute information of the seventh detection box = (1 - fusion weight) * attribute information of the first historical tracked obstacle + fusion weight * attribute information of the first part. In practical applications, the eighth overlap degree between the second part and the second historical tracked obstacle in the tracked obstacle list is calculated. If the eighth overlap degree is greater than a set threshold, it indicates that the second historical tracked obstacle and the second part are the same obstacle. In this case, the eighth detection box corresponding to the second part is determined based on the attribute information of the second part and the attribute information of the second historical tracked obstacle.
[0143] If the eighth overlap is less than a set threshold, then a fourth obstacle information is created for the second part according to the attribute information of the second part.
[0144] In practical applications, determining the eighth detection box corresponding to the second part based on the attribute information of the second part and the attribute information of the second historical tracked obstacles includes: calculating the confidence score of the second part based on the attribute information of the second part; calculating the confidence score of the second historical tracked obstacles based on the attribute information of the second historical tracked obstacles; and calculating the fusion weight based on the confidence scores of the second part and the second historical tracked obstacles, where the fusion weight = confidence score of the second part / (confidence score of the second part + confidence score of the second historical tracked obstacles).
[0145] Based on the fusion weight, the attribute information of the second part and the attribute information of the second historical tracked obstacle are combined to determine the attribute information of the eighth detection box corresponding to the second part. The attribute information of the eighth detection box = (1 - fusion weight) * attribute information of the second historical tracked obstacle + fusion weight * attribute information of the second part.
[0146] Matching obstacles with multiple occupied grid clusters revealed that no obstacle matched any object in the occupied grid cluster. Figure 9-a As shown, after processing using the methods in steps 806 and 810, the information on the third obstacle and the fourth obstacle is obtained, as follows: Figure 9-bAs shown, 901 represents the third obstacle information and 902 represents the fourth obstacle information. The obstacle detection boxes corresponding to the third obstacle information and the obstacle detection boxes corresponding to the fourth obstacle information are used as tracking obstacles.
[0147] The overlap threshold can be determined based on different occupied grid cluster objects. Different occupied grid cluster objects can be set with the same overlap threshold or different overlap thresholds. This application does not impose specific restrictions on this.
[0148] In this embodiment, the accuracy of obstacle detection is improved by optimizing the tracking of obstacles.
[0149] Example 6 refer to Figure 10 The diagram shows a structural diagram of an obstacle detection device according to an embodiment of this application. The device includes: The acquisition module 1001 is used to acquire obstacle information around the vehicle and the occupied grid clustering objects corresponding to the obstacle information. The first judgment module 1002 is used to determine whether the object occupying the grid cluster is a large vehicle.
[0150] The parameter module 1003 is used to obtain the feature parameters of the occupied grid cluster object if the occupied grid cluster object is a large vehicle and is in a turning scene. The feature parameters include: the turning point of the occupied grid cluster object and the longest axis of the occupied grid cluster object.
[0151] The splitting module 1004 is used to split the occupied grid clustered object according to the feature parameters to obtain a first part and a second part.
[0152] The fusion module 1005 is used to fuse the first part and the second part to obtain the tracked obstacle.
[0153] Optionally, the parameter module determines that the occupying grid clustered object is in a turning scene by: The first identification unit is used to identify the longest axis of the clustered objects occupying the grid; The second identification unit is used to identify the initial candidate bend points of the occupied grid clustered objects; The first calculation unit is used to calculate the overall confidence level of the initial candidate bending points and determine the core candidate points of the occupying grid clustered objects according to the ranking of the overall confidence level. The processing unit is used to process the core candidate points according to the set rules to obtain the bending points of the objects occupying the grid cluster; The output unit is used to determine that the occupying grid cluster is in a turning scene if the distance from the bending point to the longest axis of the occupying grid cluster is greater than a set threshold.
[0154] Optionally, the processing unit includes: The first judgment submodule is used to determine whether the overall confidence level of each core candidate point is greater than the confidence threshold; The first output submodule is used to perform a weighted average of the core candidate points whose overall confidence level is greater than the confidence level threshold, so as to obtain the bending point.
[0155] The first output submodule is further configured to, if the overall confidence scores of the core candidate points are all less than the confidence score threshold, take the grid farthest from the longest axis of the occupied grid cluster object as the bending point.
[0156] Optionally, the splitting module includes: A traversal unit is used to traverse all occupied grids in the occupied grid cluster object; The second calculation unit is used to calculate the first vector from the bending point to each of the all occupied grids; The third calculation unit is used to calculate the dot product of the direction vector of the longest axis and the first vector, wherein the longest axis is the line connecting the two furthest grids in the occupied grid cluster object; The first part unit is used to determine that the current grid belongs to the first direction of the longest axis if the dot product result is greater than zero, and to take all grids belonging to one side of the first direction as the first part; The second part unit is used to determine that the current grid belongs to the second direction of the longest axis if the dot product result is less than zero, and to take all grids belonging to one side of the second direction as the second part.
[0157] Optionally, the fusion module includes: The first matching unit is used to match the obstacle information with the occupied grid clustered objects to obtain an obstacle information matching the occupied grid clustered objects; The first overlap unit is used to calculate the first overlap between the obstacle information and the first part. If the first overlap is greater than a set overlap threshold, it is determined that the first part and the obstacle information correspond to the same obstacle. The first detection unit is used to determine the first detection box corresponding to the first part based on the attribute information of the obstacle information and the attribute information of the first part. The second overlap unit is used to calculate the second overlap between the obstacle information and the second part. If the second overlap is less than the set overlap threshold, the attribute information of the second part is used as the second detection box corresponding to the second part. The first fusion unit is used to obtain the tracked obstacle based on the first detection frame and the second detection frame.
[0158] Optionally, the fusion module includes: The second matching unit is used to match the obstacle information with the occupied grid clustered objects to obtain two obstacle information matching the occupied grid clustered objects, the two obstacle information being the first obstacle information and the second obstacle information, respectively. The third overlap unit is used to calculate the third overlap between the first obstacle information and the first part. If the third overlap is greater than a set threshold, it is determined that the first obstacle information and the first part correspond to the same obstacle. The second detection unit is used to determine the third detection box corresponding to the first part based on the attribute information of the first obstacle information and the attribute information of the first part. The fourth overlap unit is used to calculate the fourth overlap between the first obstacle and the second part if the third overlap is less than a set threshold; and to determine that the first obstacle information and the second part correspond to the same obstacle if the fourth overlap is greater than the set overlap threshold. The second detection unit is used to determine the fourth detection box corresponding to the second part based on the attribute information of the first obstacle information and the attribute information of the second part. The fifth overlap unit is used to calculate the fifth overlap between the second obstacle information and the first part. If the fifth overlap is greater than a set overlap threshold, it is determined that the second obstacle information and the first part correspond to the same obstacle. The third detection unit is used to determine the fifth detection box corresponding to the first part based on the attribute information of the second obstacle information and the attribute information of the first part. The sixth overlap unit is used to calculate the sixth overlap between the second obstacle and the second part if the fifth overlap is less than a set threshold; and to determine that the second obstacle information and the second part correspond to the same obstacle if the sixth overlap is greater than the set overlap threshold. The fourth detection unit is used to determine the sixth detection box corresponding to the second part based on the attribute information of the second obstacle information and the attribute information of the second part; The second fusion unit is used to obtain the tracked obstacle based on the third detection frame and the sixth detection frame or based on the fourth detection frame and the fifth detection frame.
[0159] Optionally, the fusion module includes: The third matching unit is used to match the obstacle information with the occupied grid clustered objects to obtain no obstacle information that matches the occupied grid clustered objects; The seventh overlap unit is used to calculate the seventh overlap between the first part and the historical tracked obstacles in the tracked obstacle list; A new unit is created to create a third obstacle information for the first part according to the attribute information of the first part if the seventh overlap is less than a set overlap threshold. The eighth overlap unit is used to calculate the eighth overlap between the second part and the historical tracked obstacles in the tracked obstacle list; The newly created unit is also used to create a fourth obstacle information for the second part according to the attribute information of the second part if the eighth overlap is less than a set overlap threshold. The third fusion unit is used to obtain the tracked obstacle based on the third obstacle information and the fourth obstacle information.
[0160] In this embodiment, the occupies grid cluster objects are split into a first part and a second part by using the feature parameters of the occupies grid cluster objects. The first part and the second part are fused to obtain the tracked obstacle, thereby enabling accurate detection of unknown types of obstacles to avoid obstacles. This significantly improves the accuracy and stability of target detection in complex scenes. At the same time, by determining whether the target obstacle is a large vehicle, the turning scene of a large vehicle can be detected.
[0161] This application also provides an electronic device 110, please refer to... Figure 11 It includes a processor 1110 and a memory 1120, wherein the memory 1120 is used to store computer programs; the processor 1110 is used to execute the programs stored in the memory 1120 to implement an obstacle detection method described in any embodiment of this application.
[0162] This application also provides a vehicle that stores a computer program, which, when executed by a processor, implements an obstacle detection method described in any embodiment of this application.
[0163] In this application, "multiple" refers to two or more.
[0164] The terms “first,” “second,” “third,” “fourth,” etc., used in this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.
[0165] In this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, in this application, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.
[0166] Unless otherwise specified, all steps in this application may be performed sequentially or randomly. For example, if the method includes steps A and B, it means that the method may include steps A and B performed sequentially, or it may include steps B and A performed sequentially. For example, if the method may also include step C, it means that step C may be added to the method in any order. For example, the method may include steps A, B, and C, or it may include steps A, C, and B, or it may include steps C, A, and B, etc.
[0167] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. An obstacle detection method, characterized in that, include: Obtain information about obstacles around the vehicle and the corresponding occupancy grid clustered objects; Determine whether the object occupying the grid cluster is a large vehicle; If the occupied grid clustered object is a large vehicle and is in a turning scene, the feature parameters of the occupied grid clustered object are obtained. The feature parameters include: the turning point of the occupied grid clustered object and the longest axis of the occupied grid clustered object. The occupied grid clustered objects are split according to the feature parameters to obtain a first part and a second part; The first part and the second part are fused together to obtain the tracked obstacle.
2. The method according to claim 1, characterized in that, The following method is used to determine that the object occupying the grid cluster is in a turning scene: Identify the longest axis of the occupies grid clustered objects; Identify the initial candidate bend points of the occupied grid clustered objects; Calculate the overall confidence level of the initial candidate bending points; The core candidate points of the occupied grid cluster objects are determined by ranking them according to their overall confidence level. The core candidate points are processed according to the set rules to obtain the bending points of the objects occupying the grid cluster; If the distance from the bend point to the longest axis of the occupied grid cluster object is greater than a set threshold, then the occupied grid cluster object is determined to be in a turning scene.
3. The method according to claim 2, characterized in that, The process of processing the core candidate points according to the set rules to obtain the bending points of the objects occupying the grid cluster includes: Determine whether the overall confidence level of each core candidate point is greater than the confidence threshold; If the overall confidence level of the core candidate points is greater than the confidence level threshold, then the core candidate points with overall confidence levels greater than the confidence level threshold are weighted by overall confidence level to obtain the bending point; If the overall confidence scores of the core candidate points are all less than the confidence threshold, then the grid cell furthest from the longest axis of the occupied grid cluster object is taken as the bending point.
4. The method according to claim 1, characterized in that, The step of splitting the clustered objects occupying the grid according to the feature parameters to obtain the first part and the second part includes: Iterate through all occupied graticles in the occupied graticle cluster object; Calculate the first vector from the bend point to each of the all occupied grids; Calculate the dot product of the direction vector of the longest axis and the first vector, where the longest axis is the line connecting the two furthest grid cells in the occupied grid cluster object; If the dot product result is greater than zero, then the current grid cell is determined to belong to the first direction of the longest axis, and all grid cells belonging to one side of the first direction are taken as the first part; If the dot product result is less than zero, then the current grid cell is determined to belong to the second direction of the longest axis, and all grid cells belonging to one side of the second direction are taken as the second part.
5. The method according to claim 1, characterized in that, The first part and the second part are fused together to obtain the tracked obstacles, including: The obstacle information is matched with the occupied grid clustered objects to obtain a set of obstacle information that matches the occupied grid clustered objects; Calculate the first overlap between the obstacle information and the first part. If the first overlap is greater than a set overlap threshold, then determine that the first part and the obstacle information correspond to the same obstacle. The first detection box corresponding to the first part is determined based on the attribute information of the obstacle information and the attribute information of the first part; Calculate the second overlap between the obstacle information and the second part. If the second overlap is less than a set overlap threshold, then use the attribute information of the second part as the second detection box corresponding to the second part. The tracking obstacle is obtained based on the first detection box and the second detection box.
6. The method according to claim 1, characterized in that, The first part and the second part are fused together to obtain the tracked obstacles, including: The obstacle information is matched with the occupied grid clustered objects to obtain two obstacle information matching the occupied grid clustered objects, namely the first obstacle information and the second obstacle information. Calculate the third overlap between the first obstacle information and the first part. If the third overlap is greater than a set overlap threshold, then determine that the first obstacle information and the first part correspond to the same obstacle. The third detection box corresponding to the first part is determined based on the attribute information of the first obstacle information and the attribute information of the first part. If the third overlap is less than a set threshold, then the fourth overlap between the first obstacle and the second part is calculated; if the fourth overlap is greater than the set overlap threshold, then it is determined that the first obstacle information and the second part correspond to the same obstacle. The fourth detection box corresponding to the second part is determined based on the attribute information of the first obstacle information and the attribute information of the second part; Calculate the fifth overlap between the second obstacle information and the first part. If the fifth overlap is greater than a set overlap threshold, then determine that the second obstacle information and the first part correspond to the same obstacle. The fifth detection box corresponding to the first part is determined based on the attribute information of the second obstacle information and the attribute information of the first part; If the fifth overlap is less than a set threshold, then the sixth overlap between the second obstacle and the second part is calculated; if the sixth overlap is greater than the set overlap threshold, then it is determined that the second obstacle information and the second part correspond to the same obstacle. The sixth detection box corresponding to the second part is determined based on the attribute information of the second obstacle information and the attribute information of the second part; The obstacle to be tracked is obtained based on the third and sixth detection boxes or based on the fourth and fifth detection boxes.
7. The method according to claim 1, characterized in that, The first part and the second part are fused together to obtain the tracked obstacles, including: The obstacle information is matched with the occupied grid cluster objects, and it is found that no obstacle information matches the occupied grid cluster objects; Calculate the seventh overlap between the first part and the historical tracked obstacles in the tracked obstacle list; If the seventh overlap is less than a set threshold, then a third obstacle information is created for the first part according to the attribute information of the first part; Calculate the eighth overlap between the second part and the historical tracked obstacles in the tracked obstacle list; If the eighth overlap is less than a set threshold, then a fourth obstacle information is created for the second part according to the attribute information of the second part; The tracking obstacle is obtained based on the third obstacle information and the fourth obstacle information.
8. An obstacle detection device, characterized in that, include: The acquisition module is used to acquire obstacle information around the vehicle and the occupancy grid clustering objects corresponding to the obstacle information; The first judgment module is used to determine whether the object occupying the grid cluster is a large vehicle; The parameter module is used to obtain the feature parameters of the occupying grid cluster if the occupying grid cluster is a large vehicle and is in a turning scene. The feature parameters include: the turning point of the occupying grid cluster and the longest axis of the occupying grid cluster. The splitting module is used to split the occupied grid clustered object according to the feature parameters to obtain a first part and a second part; The fusion module is used to fuse the first part and the second part to obtain the tracked obstacle.
9. An electronic device, characterized in that, Including processor and memory, among which Memory, used to store computer programs; A processor, when executing a computer program stored in a memory, implements an obstacle detection method according to any one of claims 1-7.
10. A vehicle, characterized in that, The vehicle contains a computer program that, when executed by a processor, implements an obstacle detection method according to any one of claims 1-7.