Method, system and storage medium for detecting vehicle passable area
By combining point cloud clustering and occupancy grid methods, and using lifecycle values to update clusters and calculate boundaries, the balance problem between detection accuracy and computational consumption in existing technologies is solved, and high-precision and low-consumption vehicle-accessible area detection is achieved.
Patent Information
- Application Number
- CN202211071917.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-09-01
- Publication Date
- 2025-09-30
- Estimated Expiration
- 2042-09-01
AI Technical Summary
Existing methods for detecting vehicle-accessible areas struggle to strike a balance between accuracy and computational overhead. Point cloud clustering-based methods are fast but have low accuracy, while occupancy grid-based methods are highly accurate but computationally complex and consume a lot of memory, especially at high resolutions where the amount of computation increases exponentially.
Combining point cloud clustering and occupancy grid method, the lifetime value of the previous frame is introduced into point cloud clustering to update the cluster, and the cluster boundary is calculated in the occupancy grid method, which reduces the computational cost and improves the detection accuracy.
It achieves high-precision detection of vehicle-travable areas with low computational consumption, fast detection speed, and adaptability to complex environmental changes.
Smart Images
Figure CN115331189B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent driving passable area detection, and in particular to a vehicle passable area detection method, system and storage medium. Background Art
[0002] In intelligent driving, it is necessary to detect vehicle-accessible areas. Vehicle-accessible areas are areas that vehicles can reach without being blocked by obstacles.
[0003] Currently, millimeter-wave radar is often used to detect vehicle-passable areas. There are generally two types of detection methods: point cloud clustering-based methods and occupancy grid-based methods. In the point cloud clustering-based method, 1) the millimeter-wave radar acquires point cloud data; 2) the point cloud data is clustered to form one or more clusters; 3) the clusters are mapped to a spatial area centered on the vehicle to obtain an obstacle distribution map, thereby generating a vehicle-passable area. In the occupancy grid-based method, the grid map corresponds to the space in the direction of the vehicle's travel. The grid map has a fixed resolution and size, and the free or occupied state of each grid is represented by probability; 1) the millimeter-wave radar acquires point cloud data, 2) several frames of the point cloud data are projected onto the grid map according to coordinates, and the free or occupied state of each grid is confirmed based on a probability threshold; 3) based on all the grids in the free state, the vehicle-passable area is generated.
[0004] The point cloud clustering method only targets the point cloud data of the current frame and has a fast calculation speed. The occupancy grid method combines the point cloud data of historical frames and has high detection accuracy, but the calculation is complex, consumes a lot of memory, and takes a long time to calculate. Moreover, as the resolution or size of the map increases, the calculation cost increases exponentially.
[0005] Therefore, technicians in this field are committed to developing a vehicle-traversable area detection method, system and storage medium that can combine the advantages of the two methods, with high detection accuracy, low computational memory consumption and fast detection speed. Summary of the Invention
[0006] In view of the above-mentioned defects of the prior art, the present invention provides a vehicle-traversable area detection method, system and storage medium, which can combine the advantages of point cloud clustering and occupancy grid methods, with high detection accuracy, low computational memory consumption and fast detection speed.
[0007] Specifically, the technical solution of the present invention is as follows:
[0008] In one aspect, the present invention provides a method for detecting a vehicle-travelable area, comprising the following steps:
[0009] Step 1: Update the coordinate system based on the vehicle's posture, and the millimeter-wave radar obtains the initial point cloud of the current frame;
[0010] Step 2: Filter the initial point cloud to obtain a second point cloud;
[0011] Step 3: Based on the second cluster of the second point cloud of the previous frame, the second point cloud of the current frame is initially clustered to obtain the first cluster of the current frame, and the life cycle value of the first cluster of the current frame is calculated; based on the life cycle value of the first cluster of the current frame, the first cluster of the current frame is updated to obtain the second cluster of the current frame;
[0012] Step 4: Project the second cluster of the current frame onto the grid map to obtain the boundary of the second cluster of the current frame;
[0013] Step 5: Based on the boundary calculation results, determine the area where vehicles can pass.
[0014] Furthermore, the filtering in step 2 includes clearing static points within the setting range of the moving points; wherein the moving points are set as reflection points of targets moving relative to the ground, and the static points are set as reflection points of targets stationary relative to the ground.
[0015] Furthermore, the filtering in step 2 includes removing points whose distance from the vehicle is greater than a set distance.
[0016] Furthermore, the filtering in step 2 includes removing points whose reflection intensity is lower than a set intensity.
[0017] Furthermore, the filtering in step 2 also includes clearing points that exceed a set storage quantity and retaining N points closest to the vehicle; wherein N is the set storage quantity.
[0018] Furthermore, in step 4, the boundary is found using the bidirectional index connectivity method.
[0019] Furthermore, step 5 of determining the vehicle's traversable area specifically includes: dividing the plane centered on the vehicle into several sectors according to a set step angle; mapping the boundary of each cluster to the corresponding sector; selecting the closest point from each sector to the vehicle; and smoothing each closest point to obtain the traversable area.
[0020] Furthermore, in step 5, a B-spline curve method is used for smoothing.
[0021] In a second aspect, the present invention also provides a vehicle passable area detection system, including a millimeter wave radar; the millimeter wave radar includes a point cloud acquisition unit, a point cloud clustering unit, a boundary calculation unit and a passable area calculation unit connected in sequence; the point cloud acquisition unit is configured to acquire the initial point cloud of the current frame based on the current coordinate system, and filter the initial point cloud to obtain a second point cloud, and then transmit the second point cloud to the point cloud clustering unit; the point cloud clustering unit is configured to initially cluster the second point cloud of the current frame based on the second cluster of the second point cloud of the previous frame to obtain a first cluster of the current frame, and calculate the life cycle value of the first cluster of the current frame; based on the life cycle value of the first cluster of the current frame, update the first cluster of the current frame to obtain the second cluster of the current frame; the boundary calculation unit is configured to calculate the boundary of the second cluster of the current frame using the occupancy grid method based on the data transmitted by the point cloud clustering unit; the passable area calculation unit is configured to determine the vehicle passable area based on the data transmitted by the boundary calculation unit.
[0022] In a third aspect, the present invention further provides a storage medium storing instructions, wherein the instructions are executed by one or more processors to implement the steps of the aforementioned vehicle passable area detection method.
[0023] Compared with the existing technology, the present invention combines point cloud clustering with the grid occupation method, introduces the life cycle value of the previous frame in the point cloud clustering process, updates the clusters after clustering, and has high accuracy; and uses the grid occupation method to calculate the cluster boundaries of the current frame, reducing the amount of calculation. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] The preferred embodiments will be described below in a clear and understandable manner with reference to the accompanying drawings to further illustrate the above-mentioned characteristics, technical features, advantages and implementation methods of the present invention.
[0025] Figure 1 is a flow chart of a vehicle passable area detection method according to an embodiment of the present invention;
[0026] Figure 2 Schematic diagram of the relative positions of the vehicle and the obstacle (person) in the previous frame and the current frame according to an embodiment of the present invention;
[0027] Figure 3 Schematic diagram of a vehicle passable area before and after smoothing according to an embodiment of the present invention. DETAILED DESCRIPTION
[0028] The following describes several preferred embodiments of the present invention with reference to the accompanying drawings to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms of embodiments, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.
[0029] In this application, a point cloud refers to a collection of points. Each point contains attributes including, but not limited to, one or more of the reflection point's spatial coordinates, color (R / G / B), intensity, and category. Point clouds can be obtained through various radar scans or multi-viewing algorithms.
[0030] Example 1
[0031] Figure 1 FIG. 1 is a flow chart of a method for detecting a vehicle passable area according to the present invention, wherein the method comprises the following steps:
[0032] Step 1: Update the coordinate system based on the vehicle's posture, and the millimeter-wave radar obtains the initial point cloud of the current frame;
[0033] Step 2: Filter the initial point cloud to obtain a second point cloud;
[0034] Step 3: Based on the second cluster of the second point cloud of the previous frame, the second point cloud of the current frame is initially clustered to obtain the first cluster of the current frame, and the life cycle value of the first cluster of the current frame is calculated; based on the life cycle value of the first cluster of the current frame, the first cluster of the current frame is updated to obtain the second cluster of the current frame;
[0035] Step 4: Project the second cluster of the current frame onto the grid map to obtain the boundary of the second cluster of the current frame;
[0036] Step 5: Based on the boundary calculation results of step 4, determine the vehicle passable area.
[0037] Because the posture of the vehicle is constantly changing during driving, it is necessary to calculate the point cloud information based on the current coordinate system. Figure 2 The following is an example: Figure 2 The left side shows the relative positions of the ego vehicle and the obstacle (person) in the previous frame, and the right side shows the relative positions of the ego vehicle and the obstacle (person) in the current frame. While the pose and position of the obstacle (person) remain unchanged between the current and previous frames, the ego vehicle has moved forward a certain distance and its direction of travel has shifted by an angle θ. This indicates that the relative positions of the same obstacle and the millimeter-wave radar change at different times, resulting in different point cloud information. Therefore, the millimeter-wave radar must calculate the initial point cloud information based on the current coordinate system.
[0038] Considering the large amount of noise around the driving environment, the captured initial point cloud must be filtered.
[0039] In some implementation methods, the filtering of step 2 includes but is not limited to 1) clearing static points within a set range of moving points; wherein the moving points are set as reflection points of targets moving relative to the ground, and the static points are set as reflection points of targets stationary relative to the ground; 2) clearing points whose distance from the vehicle is greater than a set distance; 3) clearing points whose reflection intensity is lower than a set intensity.
[0040] Considering the limited storage space of millimeter-wave radar, the number of points stored in the captured initial point cloud needs to be limited.
[0041] In some implementations, the captured initial point cloud removes points that exceed a set storage number and retains N points closest to the vehicle; where N is the set storage number.
[0042] Step 3 clusters the second point cloud of the current frame into clusters. Point cloud clustering involves dividing a dataset into clusters based on a specific criterion, ensuring that data within the same cluster is as similar as possible and data in different clusters are as dissimilar as possible. Clustering criteria include, but are not limited to, distance, normal vector, density, elevation, and intensity.
[0043] This example uses DBSCAN (Density-Based Spatial Clustering of Applications with Noise) as an example for clustering. The DBSCAN algorithm can find clusters of any shape in point cloud data and can also find noise points that do not belong to any cluster. Specifically, the DBSCAN clustering method includes the following steps:
[0044] Step S1, set the cluster radius r, and set the minimum number of samples minpts contained in the radius r;
[0045] Step S2: In the second point cloud, record the number of adjacent points within the radius r of each point; if the number of adjacent points is greater than or equal to the minimum sample size minpts, the point is set as a core point;
[0046] Step S3: cluster the core points into clusters, so that: the distance between any core points in the cluster is greater than the radius r; the distance within the cluster includes the core point and other points within the radius r; if a core point in the cluster has one or more adjacent core points, at least one of the distances between the core point and the one or more adjacent core points is less than or equal to the radius r;
[0047] Step S4: Repeat step S3 until all points in the second point cloud are clustered.
[0048] In some embodiments, step S3 specifically includes the following steps:
[0049] Randomly select a core point p, cluster the core point p and the points it covers within radius r into a cluster c; traverse all the adjacent points of the core point p within radius r, and if there is a core point q, cluster the new points covered by the core point q within radius r into cluster c; and so on, until cluster c no longer adds new core points, and the clustering is completed.
[0050] Traditional point cloud clustering methods only process the current frame and are computationally fast. Traditional grid occupancy methods, on the one hand, rely on historical frame information, and on probabilistic calculations, combining these two to calculate obstacle boundaries with high accuracy. However, grid occupancy methods are computationally intensive, and this computational complexity increases exponentially with increasing grid map resolution and size. Therefore, in this embodiment, point cloud clustering updates clusters based on historical frame information, while the grid occupancy method calculates the boundaries of the updated clusters for the current frame.
[0051] Due to changes in the vehicle's posture, the features of the same cluster in different frames may be completely different. Therefore, it is necessary to establish a connection between the historical frame and the current frame in order to update the clustering results of the current frame based on the historical frame information.
[0052] In some implementations, step 3 specifically includes the following steps:
[0053] Step 3.1: In the second point cloud of the current frame, record the number of points within a given range for each point; if the number is greater than or equal to a first threshold, set the point as a core point;
[0054] Step 3.2: Based on the second cluster of the second point cloud of the previous frame, perform the initial clustering on the second point cloud of the current frame to obtain the first cluster of the current frame, and calculate the life cycle value of the first cluster of the current frame;
[0055] The life cycle value includes a continuous loss value; if the first cluster of the current frame contains one or more core points of the current frame, the continuous loss value is set to zero; otherwise, the continuous loss value is increased by 1;
[0056] Step 3.3: If the continuous loss value of the first cluster of the current frame is greater than or equal to the continuous loss fixed value, delete the cluster;
[0057] Step 3.4: Repeat steps 3.2 to 3.4 for the remaining first clusters of the current frame.
[0058] The purpose of setting core points is to make the point clouds of different frames have connections that can be segmented or clustered, so that the cluster of the current frame can update the point cloud clustering results based on the information of the historical frames; secondly, the core points can more concentratedly reflect the characteristics of the cluster.
[0059] In other implementations, it is also possible to choose not to set core points. That is, step 3.1 is canceled, and step 3.2 includes object discrimination, directly using whether the second point cloud of the current frame is newly added as the discrimination criterion.
[0060] The accumulation process of continuous loss values being less than the continuous loss constant slows down the "disappearance" speed of the same cluster in different frames, which is conducive to capturing the true boundaries of irregular obstacles, is closer to the actual usage scenario, and improves the safety margin of the detection of the passable area. For example, as the driving posture of the vehicle changes, a continuously discontinuously distributed low-reflection intensity obstacle may be identified as an obstacle at time A, identified as a passable area at time A+1, and identified as an obstacle again at time A+2. Through lifecycle management, the obstacle identified at time A can "continue" for one frame, that is, it will still be considered an obstacle at time A+1. Through this setting, the cylindrical obstacle can eventually be identified as one object rather than several discrete long objects.
[0061] Slowing down the disappearance of clusters also facilitates collaboration among multiple millimeter-wave radars. To improve autonomous driving detection accuracy, a single intelligent vehicle is equipped with multiple millimeter-wave radars. Due to different installation locations, different millimeter-wave radars generate different point cloud data for the same obstacle. Because data exchange or aggregation between different millimeter-wave radars takes time, collaborative work requires a high degree of synchronization. Setting the continuous loss value and the continuous loss constant is equivalent to "caching" the obstacle point cloud for a certain number of frames, which facilitates real-time data interaction and coordination.
[0062] In addition, the continuous loss value keeps increasing until it reaches a continuous loss constant value, indicating that the collected point cloud corresponds to a false obstacle, and the first cluster corresponding to the false obstacle should be deleted.
[0063] In other embodiments, the life cycle value may also include a continuous observation value; so that: if the first cluster of the current frame contains one or more core points of the current frame, the continuous observation value is increased by 1; if the continuous observation value is greater than or equal to the continuous observation value, several frames of the cluster are retained.
[0064] The continuous observation value increases until it reaches the continuous observation value, indicating that the collected point cloud corresponds to a real obstacle, and the first cluster corresponding to the real obstacle should be retained.
[0065] This setup helps distinguish temporary occlusions from actual obstacles. Specifically, a cluster is confirmed to be true only after several consecutive frames, which can filter out temporary occlusions, such as birds flying in the air or floating objects in the way.
[0066] In other implementations, depending on actual needs, the continuous loss value and the continuous observation value can be used in conjunction. For example, a continuous observation value and a continuous observation value can be set, while a continuous loss value and a continuous loss fixed value can also be set; and the continuous observation value has a higher priority than the continuous loss value. In this case, the second cluster, which has been retained for several frames, will not be deleted because the continuous loss value reaches the continuous loss fixed value.
[0067] It should also be noted that the historical frame in this embodiment is the previous frame. There is no such limitation in actual application, and several previous frames can be used as needed.
[0068] It should also be noted that during initialization, since there is no point cloud of the previous frame, the lifetime value of the previous frame is set to zero and the second cluster of the previous frame is empty.
[0069] Step 4 is to calculate the boundary of the second cluster of the current frame based on the occupancy grid method.
[0070] In some implementations, calculating only one cluster at a time allows for the use of a smaller grid map. This reduced computational effort also reduces the performance requirements of the computing module, thereby lowering the cost of the vehicle radar processor. Furthermore, given the varying shapes of obstacles corresponding to each cluster, the size and resolution of the grid map can be dynamically adjusted based on the size of each cluster.
[0071] In some implementations, the size and resolution of the grid map are calculated as follows:
[0072]
[0073] Where gridRes is the grid map resolution, gridNum is the resolution adjustment constant, boxSizeX is the size of the cluster in the direction of the vehicle's travel, and boxSizeY is the size of the cluster perpendicular to the vehicle's travel. boxSizeX and boxSizeY are calculated based on the cluster size; in some implementations, they are calculated based on the X and Y coordinates of the point cloud within the cluster.
[0074] In some implementations, downsampling can be performed when each grid contains multiple points. During downsampling, the points in the grid that best represent the probability of the grid being occupied are selected, which does not affect the grid state determination while reducing the computational effort.
[0075] In this embodiment, step 4 specifically includes the following steps:
[0076] Step 4.1, calculating the size and resolution of the grid map based on the second cluster of the current frame;
[0077] Step 4.2: Project the second cluster of the current frame onto the grid map and count the grid features; the features include the number of points in each grid and the coordinate information of the occupied grids;
[0078] Step 4.3: Calculate the cluster boundaries.
[0079] Step 4.4: Repeat steps 4.1 to 4.3 until the boundaries of all second clusters in the current frame are calculated.
[0080] Step 4.3 can use a variety of methods to calculate the cluster boundaries. In some implementations, a bidirectional index connectivity method is used to calculate the cluster boundaries, which is simple in principle and highly accurate.
[0081] This embodiment calculates each cluster sequentially, reducing the performance requirements of the computing module. In practical scenarios, step 4 can also be used to calculate the boundary of the second cluster in the current frame in parallel. Parallel processing can more quickly calculate cluster boundaries, facilitating multi-radar coordination. Alternatively, parallel computing chips such as ARM-based multi-core processors, x86-based processors, or FPGA-based processors can be used.
[0082] Step 4 calculates the boundary of the obstacle. Step 5 determines the vehicle's passable area based on the calculation result of the obstacle boundary.
[0083] In some implementations, step 5 specifically includes the following steps:
[0084] Step 5.1: Divide the plane centered on the vehicle into several sectors according to the set step angle;
[0085] Step 5.2: Map the boundaries of each cluster to the corresponding sector; select the closest point of each sector to the vehicle;
[0086] Step 5.3: Smooth each nearest point to obtain the passable area.
[0087] The distribution of the closest points of each cluster to the vehicle in the sector may be relatively discrete, such as Figure 3 As shown on the left side of the figure. For practical driving, a passable space smaller than a certain size is meaningless. Therefore, smoothing is required for each nearest point.
[0088] In some implementations, a B-spline curve method is used for smoothing, which has a simple principle and high accuracy.
[0089] Figure 3 The right side shows the schematic diagram of the traversable area after smoothing. Figure 3 It can be seen that after smoothing, the boundary of the passable area that can actually be passed can be obtained.
[0090] Compared with the existing technology, this embodiment combines point cloud clustering with the occupancy grid method, introduces the life cycle value of the previous frame in the point cloud clustering process, updates the clusters after clustering, and has high accuracy; the occupancy grid method is used to calculate the boundaries of the clusters of the current frame, reducing the amount of calculation.
[0091] Example 2
[0092] The present invention also provides a vehicle passable area detection system.
[0093] The detection system includes a millimeter-wave radar; the millimeter-wave radar includes a point cloud acquisition unit, a point cloud clustering unit, a boundary calculation unit and a passable area calculation unit connected in sequence; the point cloud acquisition unit is configured to acquire the initial point cloud of the current frame based on the current coordinate system, filter the initial point cloud to obtain a second point cloud, and then transmit the second point cloud to the point cloud clustering unit; the point cloud clustering unit is configured to initially cluster the second point cloud of the current frame based on the second cluster of the second point cloud of the previous frame to obtain a first cluster of the current frame, and calculate the life cycle value of the first cluster of the current frame; based on the life cycle value of the first cluster of the current frame, the first cluster of the current frame is updated to obtain the second cluster of the current frame; the boundary calculation unit is configured to calculate the boundary of the cluster using the occupancy grid method based on the data transmitted by the point cloud clustering unit; the passable area calculation unit is configured to determine the vehicle passable area based on the data transmitted by the boundary calculation unit.
[0094] In some implementation methods, the millimeter wave radar can be a 4D millimeter wave radar, which can directly obtain the distance, direction, speed and height information of the obstacle point cloud.
[0095] Example 3
[0096] The present invention also provides a storage medium storing instructions, wherein the instructions are executed by one or more processors to implement the steps of the vehicle passable area detection method of embodiment 1.
[0097] The system embodiment of the present invention corresponds to the method embodiment of the present invention. The technical details of the method embodiment of the present invention are also applicable to the system embodiment of the present invention. To reduce repetition, they are not described again.
[0098] The preferred embodiments of the present invention have been described in detail above. It should be understood that numerous modifications and variations based on the concepts of the present invention are possible without inventive effort by those skilled in the art. Therefore, any technical solution that can be derived by one skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.
Claims
1. A method for detecting a vehicle passable area, characterized in that: The method comprises the following steps: Step 1: Update the coordinate system based on the vehicle's posture, and the millimeter-wave radar obtains the initial point cloud of the current frame; Step 2: Filter the initial point cloud to obtain a second point cloud; Step 3: Based on the second cluster of the second point cloud of the previous frame, the second point cloud of the current frame is initially clustered to obtain the first cluster of the current frame, and the life cycle value of the first cluster of the current frame is calculated; based on the life cycle value of the first cluster of the current frame, the first cluster of the current frame is updated to obtain the second cluster of the current frame; the life cycle value includes a continuous loss value; if the first cluster of the current frame contains one or more core points of the current frame, the continuous loss value is set to zero; otherwise, the continuous loss value is increased by 1; Step 4: Project the second cluster of the current frame onto the grid map to obtain the boundary of the second cluster of the current frame; Step 5: Based on the boundary calculation results, determine the area where vehicles can pass.
2. The vehicle passable area detection method according to claim 1, characterized in that: The filtering in step 2 includes clearing static points within the range of the moving point setting; wherein the moving point is set as the reflection point of the target moving relative to the ground, and the static point is set as the reflection point of the target stationary relative to the ground.
3. The vehicle passable area detection method according to claim 1, characterized in that: The filtering in step 2 includes removing points whose distance from the vehicle is greater than a set distance.
4. The vehicle passable area detection method according to claim 1, characterized in that: The filtering in step 2 includes removing points whose reflection intensity is lower than a set intensity.
5. The vehicle passable area detection method according to claim 1, characterized in that: The filtering in step 2 further includes clearing points exceeding a set storage quantity and retaining N points closest to the vehicle; wherein N is the set storage quantity.
6. The vehicle passable area detection method according to claim 1, characterized in that: In step 4, the boundary is found using a bidirectional index connectivity method.
7. The vehicle passable area detection method according to claim 1, characterized in that: The step 5 of determining the vehicle's traversable area specifically includes: dividing the plane centered on the vehicle into several sectors according to a set step angle; mapping the boundary of each cluster to the corresponding sector; selecting the closest point from each sector to the vehicle; and smoothing each closest point to obtain the traversable area.
8. The vehicle passable area detection method according to claim 7, characterized in that: The smoothing process is performed using the B-spline curve method.
9. A vehicle passable area detection system, comprising a millimeter wave radar, characterized in that: The millimeter wave radar includes a point cloud acquisition unit, a point cloud clustering unit, a boundary calculation unit and a passable area calculation unit which are connected in sequence; The point cloud acquisition unit is configured to acquire an initial point cloud of a current frame based on a current coordinate system, filter the initial point cloud to obtain a second point cloud, and then transmit the second point cloud to the point cloud clustering unit; The point cloud clustering unit is configured to initially cluster the second point cloud of the current frame based on the second cluster of the second point cloud of the previous frame to obtain the first cluster of the current frame, and calculate the life cycle value of the first cluster of the current frame; Based on the life cycle value of the first cluster of the current frame, the first cluster of the current frame is updated to obtain the second cluster of the current frame; the life cycle value includes a continuous loss value; if the first cluster of the current frame contains one or more core points of the current frame, the continuous loss value is set to zero; Otherwise, the consecutive missing values are incremented by 1; The boundary calculation unit is configured to calculate the boundary of the second cluster of the current frame using an occupancy grid method based on the data transmitted by the point cloud clustering unit; The passable area calculation unit is configured to determine a vehicle passable area based on the data transmitted by the boundary calculation unit.
10. A storage medium, characterized in that: The storage medium stores instructions, and the instructions are executed by one or more processors to implement the steps of the vehicle passable area detection method according to any one of claims 1 to 8.
Citation Information
Patent Citations
Train collision early warning method and device
CN112379393A
Automobile millimeter-wave radar plot condensation method and system
CN112731296A