An environment map-based passable area boundary extraction method

Abstract

The present application relates to a kind of based on the extraction method of passable area boundary of environmental map, belong to road boundary extraction technical field, solve the problem that the over-reliance on local geometric features in the extraction method in prior art and the continuity of boundary structure is not effectively utilized, leading to the problem that passable area boundary cannot be accurately extracted.The environmental map of integrated topological map and grid map is obtained, the historical vehicle trajectory is smoothed, the main road centerline is obtained, and then the initial passable area boundary of the main road is obtained, and then the dynamic programming algorithm is used to obtain the passable area boundary of the main road;Based on the environmental map and the main road centerline, the optimal branch of the main road is obtained, and then the passable area boundary of the optimal branch of the main road is obtained;Under the geometric constraint and kinematic constraint of vehicle, based on the passable area boundary of the main road and its each optimal branch, the passable area boundary of the main road and its each optimal passable branch is obtained.

Description

Technical Field

[0001] This invention relates to the field of road boundary extraction technology, and in particular to a method for extracting the boundaries of traversable areas based on environmental maps. Background Technology

[0002] Accessible area boundaries provide accurate environmental constraints for autonomous vehicle (RV) planning, enhancing its stability and improving real-time navigation efficiency in off-road environments. This allows RVs to more quickly generate locally planned paths that satisfy vehicle passability and smoothness based on map information such as topology maps and accessible area boundaries during real-time navigation. Therefore, accessible area boundary extraction has significant application and research value.

[0003] Most existing methods are based on extracting the boundaries of passable areas from LiDAR 3D point cloud data, identifying boundaries by analyzing point cloud density changes, height differences, or spatial geometric features (such as gradients and normal vectors). One common approach is to first extract candidate boundary points using local features such as height differences, and then combine region growing, curve fitting, or filtering techniques for boundary modeling and smoothing. Another mainstream method is to project the 3D point cloud into a raster map, and indirectly infer passable areas and their boundaries by using obstacle occupancy status and combining techniques such as ground segmentation and distance transformation.

[0004] However, most traditional offline methods for extracting accessible area boundaries extract seed points from obstacle grid maps, connect these seed points, and then extract the accessible area boundary, relying on local geometric features for boundary recognition. But in some road environments, such as off-road conditions, low rocks, shrubs, and field ridges are common static obstacles that can easily be misidentified as accessible area boundaries. This results in inaccurate extraction of the actual accessible area boundary seed points, making it difficult to accurately build an accessible area boundary model. Although the most stable property of accessible boundaries is their continuity, traditional methods mostly focus only on local features and lack the ability to model the overall continuity of the boundary.

[0005] Therefore, the main drawback of existing technologies lies in their over-reliance on local geometric features and their lack of effective utilization of the continuity of boundary structures. Summary of the Invention

[0006] In view of the above analysis, the present invention aims to provide a method for extracting the boundary of traversable areas based on environmental maps, so as to solve the problem that existing extraction methods cannot accurately extract the boundary of traversable areas due to over-reliance on local geometric features and failure to effectively utilize the continuity of boundary structure.

[0007] This invention provides a method for extracting the boundaries of traversable areas based on an environmental map, comprising the following steps:

[0008] Obtain an environment map that integrates topology and raster maps;

[0009] Based on the environmental map, the historical vehicle trajectories are smoothed to obtain the centerline of the main road, and then the initial passable area boundary of the main road is obtained. Then, a dynamic programming algorithm is used to obtain the passable area boundary of the main road.

[0010] Based on the environmental map and the center line of the main road, the optimal branch roads of the main road are obtained, and then the passable area boundaries of the optimal branch roads of the main road are obtained.

[0011] Under the geometric and kinematic constraints of the vehicle, based on the traversable area boundaries of the main road and its optimal branch roads, the traversable optimal branch roads of the main road are obtained, and thus the traversable area boundaries of the main road and its optimal branch roads are obtained.

[0012] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0013] This invention provides a method for extracting the boundaries of traversable areas based on an environmental map. The method acquires an environmental map integrating a topological map and a raster map, smooths historical vehicle trajectories to obtain the centerline of the main road, and then obtains the initial traversable area boundary of the main road. A dynamic programming algorithm is then used to obtain the traversable area boundary of the main road. Next, based on the environmental map and the centerline of the main road, the optimal branches of the main road are obtained, and the traversable area boundaries of each optimal branch are then obtained. Finally, under the geometric and kinematic constraints of the vehicles, based on the traversable area boundaries of the main road and its optimal branches, the traversable area boundaries of each branch of the main road are obtained. By identifying the optimal passable branch, the boundaries of the passable areas of the main road and its optimal passable branch can be obtained. The continuity and width distribution of the passable area boundaries can effectively reflect the continuity and shape of the road. Modeling the passable area boundary model can improve the accuracy of passable area boundary extraction, making it more consistent with the real road environment. It can be better applied to outdoor environmental scenarios and is also suitable for off-road environments with complex obstacles and irregular shapes. It can reduce the computation time for passable area boundary extraction when autonomous vehicles are navigating, and help improve the smoothness and safety of autonomous vehicle driving.

[0014] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0015] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0016] Figure 1 This is a flowchart illustrating the method for extracting the boundary of a passable area based on an environmental map, as provided in Embodiment 1 of the present invention.

[0017] Figure 2 The smoothed vehicle trajectory provided in Embodiment 1 of the present invention;

[0018] Figure 3 This is a schematic diagram of the rotation and translation of the vehicle driving coordinate system and the image coordinate system provided in Embodiment 1 of the present invention;

[0019] Figure 4 This is a schematic diagram of solving the boundary points of the passable area provided in Embodiment 1 of the present invention;

[0020] Figure 5 This is a schematic diagram of the correction of the boundary points of the passable area provided in Embodiment 1 of the present invention;

[0021] Figure 6 This is a schematic diagram of the extraction of the passable area boundary provided in Embodiment 1 of the present invention;

[0022] Figure 7 This is a schematic diagram of candidate points for the boundary of a passable area provided in Embodiment 1 of the present invention;

[0023] Figure 8 This is a probability diagram of trajectory points and passable area boundary points provided in Embodiment 1 of the present invention;

[0024] Figure 9 This is a schematic diagram of the passable area boundary obtained based on dynamic programming, provided in Embodiment 1 of the present invention.

[0025] Figure 10 This is a schematic diagram of the topology map of a typical scenario provided in Embodiment 1 of the present invention;

[0026] Figure 11 This is a schematic diagram of the results of screening the optimal branch and determining the passable endpoint provided in Embodiment 1 of the present invention;

[0027] Figure 12 This is a schematic diagram of the main road boundary and branch road boundary provided in Embodiment 1 of the present invention;

[0028] Figure 13 This is a schematic diagram of the optimal branch selection provided in Embodiment 1 of the present invention;

[0029] Figure 14 This is a schematic diagram of the connection relationship between the branch road and the vehicle's travel trajectory provided in Embodiment 1 of the present invention;

[0030] Figure 15 This is a schematic diagram of vehicle collision constraints provided in Embodiment 1 of the present invention;

[0031] Figure 16 This is a schematic diagram of a local planning path provided in Embodiment 1 of the present invention;

[0032] Figure 17 This is the truth value region of the main road area provided in Embodiment 2 of the present invention;

[0033] Figure 18 The results of the accessible area boundary extraction and branch road accessibility verification provided in Embodiment 2 of the present invention. Detailed Implementation

[0034] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0035] Example 1

[0036] A specific embodiment of the present invention discloses a method for extracting the boundaries of traversable areas based on an environmental map, such as... Figure 1 As shown, it includes the following steps:

[0037] S1. Obtain the environment map that integrates the topology map and the raster map.

[0038] Specifically, LiDAR is used for real-time localization and mapping (SLAM) to generate a grid map, and a topology map is generated based on the grid map, which together constitute the environment map. In the grid map, 0 represents no obstacles and 1 represents obstacles. In the topology map, the edges connecting vertices represent roads, vertices represent intersections, and the weights of the edges represent the travel costs, such as road width and length.

[0039] For example, in this embodiment, a grid map is used to describe the vehicle's surrounding environment, with each pixel representing an actual length of 0.2m.

[0040] S2. Based on the environmental map, the historical vehicle trajectories are smoothed to obtain the centerline of the main road, and then the initial passable area boundary of the main road is obtained. Then, a dynamic programming algorithm is used to obtain the passable area boundary of the main road.

[0041] During implementation, the initial passable area boundary of the main road is obtained in the following way:

[0042] S211. In the environmental map, the historical vehicle trajectory points are smoothed by B-spline curve fitting to obtain the center line of the main road.

[0043] Specifically, historical vehicle trajectory points are obtained through the vehicle's GPS or positioning system. This means that the paths previously traveled by the vehicle are roughly located in the center of the passable area of ​​the road. In unstructured roads (such as off-road environments), where there are no clear lane markings, the vehicle's trajectory is the most direct and reliable evidence of the road. Therefore, the centerline of the main road is determined using historical vehicle trajectory points. It can be understood that the centerline of the main road is a geometric curve.

[0044] Specifically, cubic clamped B-spline curve fitting is used to smooth each trajectory point.

[0045] Understandably, B-spline curve fitting can eliminate the jitter and abrupt changes in heading of the original historical trajectory, making it more consistent with the actual smooth motion model of the vehicle.

[0046] For example, such as Figure 2 As shown, it displays the effect of smoothing the vehicle trajectory with B-splines in a complex off-road scenario. The white dots are the original vehicle trajectory points, and the black path is the smoothed trajectory, i.e. the center line of the main road. It can be seen that the curvature of the smoothed trajectory is continuous, which better reflects the actual driving state of the vehicle and avoids sudden changes in heading.

[0047] S212. Based on the center line of the main road, obtain the normal vector of each trajectory point; based on the environment map, obtain the distance from each trajectory point to the nearest obstacle, and then obtain the road width on the left and right sides of each trajectory point.

[0048] In practice, based on the centerline of the main road, the tangent vector of each trajectory point on the main road is obtained, and the normal vector of each trajectory point is obtained based on the tangent vector of each trajectory point; wherein, the tangent vector points to the direction of vehicle movement, and the normal vector is perpendicular to the direction of movement and points to the left side of the vehicle.

[0049] Specifically, the normal vector and the tangent vector are perpendicular, and their relationship is as follows:

[0050] ,

[0051] In the formula, , These represent the components of the tangent vector on the x and y axes, respectively. , These represent the components of the normal vector on the horizontal and vertical axes, respectively.

[0052] Specifically, the tangent vector of each trajectory point is obtained by using the interpolation ratio of the horizontal and vertical coordinates of adjacent trajectory points.

[0053] In practice, the distance from each trajectory point to the nearest obstacle is obtained from the grid map in the environment map. Specifically, the distance from each pixel to the nearest obstacle and the specific coordinates of the nearest obstacle are known in the grid map, so the distance from each trajectory point to the nearest obstacle can be determined. It is understandable that the distance from the trajectory point to the nearest obstacle may be the left or right boundary of the road.

[0054] In practice, the road widths on the left and right sides of each trajectory point are obtained in the following way:

[0055] Based on each trajectory point and its corresponding nearest obstacle position, the road width to the left or right of each trajectory point is obtained, specifically:

[0056] Based on the current trajectory point and the location of its nearest obstacle, the direction parameters are obtained;

[0057] If the direction parameter is greater than 0, the nearest obstacle is determined to be to the left of the current trajectory point, and the distance from the current trajectory point to the nearest obstacle is taken as the width of the road to the left of the trajectory point; otherwise, the nearest obstacle is determined to be to the right of the current trajectory point, and the distance from the current trajectory point to the nearest obstacle is taken as the width of the road to the right of the trajectory point.

[0058] Based on the road width to the left or right of each trajectory point, linear interpolation is performed on the missing road width to the left or right of each trajectory point to obtain the road width on the left and right sides of each trajectory point.

[0059] Optionally, the direction vector is obtained by subtracting the position of the nearest obstacle from the current trajectory point, and then the direction vector and the normal vector of the current trajectory point are multiplied by a dot product to obtain the direction parameter.

[0060] Alternatively, the direction parameter can also be obtained in the following ways:

[0061] Establish the vehicle's driving coordinates with the trajectory point as the center and the tangent direction as the x-coordinate. The specific coordinates and transformation relationships are as follows: Figure 3 As shown.

[0062] Given the coordinates in the image Coordinates of the nearest obstacle point Because the vehicle's driving coordinate system rotates and translates relative to the image coordinate system, and because the vehicle's driving coordinate system rotates clockwise relative to the image coordinate system... Therefore, first establish the image coordinate system , around point Rotation We can first obtain New coordinate system mid-coordinate value The key to solving this problem is to use the known sides as hypotenuses to find the solution, combined with... Figure 3 The trigonometric functions can be used to obtain:

[0063] ,

[0064] Due to the coordinate axes Relative to coordinate system This is a translation relationship, in the image coordinates Down, The coordinates are You can first transform it into a coordinate system. From the coordinates, we can obtain:

[0065] ,

[0066] This can be obtained using vector addition:

[0067] ,

[0068] So point On the coordinate axes coordinates for:

[0069] ,

[0070] in, Indicates the direction parameter, as long as If the distance is 0, it means the nearest obstacle is on the left side of the vehicle, and the distance to the nearest obstacle is the width of the road to the left. Conversely, if the distance is 0, it is the width of the road to the right.

[0071] It should be noted that for the same trajectory point, due to its distance to the nearest obstacle, it can only belong to one side of the road width. There will inevitably be gaps in the left and right road width sequences. Therefore, linear interpolation is used for the left and right road width sequences.

[0072] S213. Each trajectory point on the center line of the main road is extended along its normal direction and based on the corresponding left and right road widths to obtain the initial passable area boundary of the main road; wherein, the passable area boundary includes the sequence of boundary points on the left and right sides of the passable area.

[0073] In practice, the left boundary of the passable area of ​​each trajectory point is the sum of the product of the left road width and the normal vector of the trajectory point, and the trajectory point itself; the right boundary of the passable area of ​​each trajectory point is the sum of the product of the right road width and the normal vector of the trajectory point, and the trajectory point itself.

[0074] For example, taking the right side as an example, such as Figure 4 As shown, trajectory points , The angle of the formed tangent is And because and Since they are complementary, we know that:

[0075] ,

[0076] In triangle Among them For trajectory points The width of the passable area to the right can be obtained using trigonometric functions:

[0077] ,

[0078] It should be noted that, under normal circumstances, the left and right boundary points of the passable area are connected sequentially, such as... Figure 5 As shown in the left part, the equations of the lines at the boundary points of adjacent passable regions do not intersect within the domain generated by the four vertices (lines). With a straight line In quadrilateral No intersections); if there are issues with the smooth boundaries of the passable area (such as...) Figure 5 As shown in the middle part, the straight line With a straight line In quadrilateral If an intersection occurs, then it is necessary to... Delete the boundary points of the two passable areas, such as Figure 5 As shown in the right part of the middle section.

[0079] Therefore, preferably, in this embodiment, the initial passable area boundary of the main road is also checked for continuity. If the left and right boundaries of the initial passable area intersect, boundary correction is performed; specifically:

[0080] Based on adjacent trajectory points, the straight line equations of the left and right boundary line segments are obtained;

[0081] Based on the line equations of the left and right boundary segments, a linear equation matrix is ​​established, and then it is determined whether the linear equation matrix is ​​of full rank.

[0082] If the rank is full, calculate the common solution and determine whether the common solution exists in the road region of adjacent trajectory points. If it exists, determine whether the left and right boundary segments of adjacent trajectory points intersect; otherwise, determine whether the left and right boundary segments of adjacent trajectory points do not intersect. The road region of adjacent trajectory points is the region enclosed by the left and right boundary points of the initial passable region of adjacent trajectory points.

[0083] Otherwise, determine that the left and right boundary line segments of adjacent trajectory points do not intersect.

[0084] Specifically, the equations of the straight lines for the left and right boundary segments are expressed as follows:

[0085] ,

[0086] In the formula, , These are the components of the abscissa of the normal vectors of adjacent trajectory points A and B, respectively. , These are the components of the ordinate of the normal vectors of adjacent trajectory points A and B, respectively. Let A be the coordinates of the initial passable area boundary point to the left of trajectory point A. Let x be the boundary point of the initially passable area to the right of trajectory point B, and let x and y be the independent and dependent variables in the equation of the line, respectively.

[0087] Specifically, the linear equation matrix is ​​represented as:

[0088] ,

[0089] Specifically, if the left and right boundary segments of adjacent trajectory points intersect, the boundary points of the passable areas on the left and right sides corresponding to the subsequent trajectory point will be deleted. Furthermore, the boundary points of the passable areas on the left and right sides of the subsequent trajectory point can be regenerated to satisfy the condition that the boundary lines do not intersect and are continuous.

[0090] It should be noted that, as Figure 6 In the left part, the vehicle's trajectory points are black, the boundary points of the passable area on the left side of the vehicle are green, and the boundary points of the passable area on the right side of the vehicle are white; for example... Figure 6 In the right-hand section, the black line segment represents the vehicle's trajectory, the green line segment represents the boundary of the passable area on the left side of the vehicle, and the white line segment represents the boundary of the passable area on the right side of the vehicle. It can be seen that the green and white passable area boundary lines can fit the actual passable area boundary quite well. However, because this method of extracting the passable area boundary is highly dependent on the position of the vehicle trajectory line, if the vehicle trajectory line is not in the middle of the road, the accuracy of the extracted passable area boundary will be greatly affected. Furthermore, because the distance from the vehicle trajectory point to the nearest obstacle is taken as half the road width, when there is an obstacle on the road, the extracted passable area boundary line will be near the obstacle and cannot reflect the actual passable area boundary. For example, at the lower left corner of the green boundary line, the passable area boundary line curves inward into the road.

[0091] Therefore, in implementation, this embodiment uses a dynamic programming algorithm to obtain the passable area boundary of the main road, including the following steps:

[0092] S221. Based on the main road centerline and its initial passable area boundary and the set width interval, generate candidate boundary point strips on the left and right sides; wherein, the candidate boundary point strips are each candidate boundary point on the side corresponding to each trajectory point;

[0093] S222. Based on the main road centerline and the candidate boundary point zone on one side, the optimal boundary point sequence on one side of the main road is obtained through dynamic programming optimization of one side boundary.

[0094] S223. Based on the optimal boundary point sequence on one side of the main road, the centerline of the main road and the candidate boundary point zone on the other side, the optimal boundary point sequence on the other side of the main road is obtained through dynamic programming optimization of the boundary on the other side.

[0095] S224. Based on the optimal boundary point sequence on both sides of the main road, the passable area boundary of the main road is obtained.

[0096] In practice, for each trajectory point on the centerline of the main road, the left and right boundary points of the passable area corresponding to each trajectory point are used to generate a set number of candidate boundary points at set width intervals along the normal direction of the corresponding trajectory point, thus generating candidate boundary point strips on the left and right sides. Figure 7 As shown.

[0097] Specifically, the width interval and the number of candidates are set according to specific needs; for example, the width interval is set to 0.2 meters and the number of candidates is set to 11.

[0098] In practice, the optimal boundary point sequences on both sides of the main road are obtained in the following ways:

[0099] Based on the main road centerline and its current side candidate boundary point zone, a probabilistic graphical model and a single-step cost function are constructed for the current side. The probabilistic graphical model is constructed based on each trajectory point on the main road centerline, with each stage's state being the candidate boundary point of the corresponding trajectory point on the current side, and the state value being the minimum cumulative cost. The minimum cumulative cost is the minimum cumulative cost to reach the current stage and select the current candidate node, obtained based on the single-step cost function. For example, such as... Figure 8 As shown, T represents the trajectory point, Y represents the candidate boundary point, and Q represents the trajectory point matching status;

[0100] Based on the constructed current-side probabilistic graphical model and single-step cost function, dynamic programming with inverse recursion is used to obtain the optimal boundary point sequence of the current side of the main road.

[0101] It should be noted that the steps for the optimal boundary point sequence on both sides of the main road are the same, only their single-step cost functions are different; and one side of the main road can be either the left or the right side, without any specific restrictions. If one side of the main road is the left side, then the other side is the right side, and similarly, if one side of the main road is the right side, then the other side is the left side.

[0102] Specifically, if the current side is the main road side, then the single-step cost function of that side includes obstacle distance cost, boundary smoothness cost, and trajectory following cost; among which, the boundary smoothness cost includes the distance cost of adjacent candidate boundary points and the road width change cost;

[0103] If the current side is the other side of the main road, then the single-step cost function of that side includes obstacle distance cost, boundary smoothness cost, trajectory following cost, and width consistency constraint cost.

[0104] More specifically, if the current side is the main road side, then the single-step cost function for that side is expressed as:

[0105] ,

[0106] In the formula, This represents the single-step cost when candidate boundary point k is selected in stage i+1 and candidate boundary point j is selected in the previous stage i. , , , These represent the weight coefficients for obstacle distance cost, trajectory following cost, distance cost between adjacent candidate boundary points, and road width variation cost, respectively. , These represent the obstacle distance cost and trajectory following cost when k is selected as the candidate boundary point k on this side in stage i+1, respectively. , These represent the distance cost and road width change cost of adjacent candidate boundary points when candidate boundary point k is selected as the candidate boundary point k on this side in stage i+1 and candidate boundary point j was selected as the candidate boundary point j on this side in the previous stage i.

[0107] For example, if the current side is the main road side, then , , , The values ​​were set to 0.4, 0.2, 0.3, and 0.1 respectively.

[0108] More specifically, if the current side is the other side of the main path, then the single-step cost function for that side is expressed as:

[0109] ,

[0110] In the formula, This represents the single-step cost when candidate boundary point k is selected in stage i+1 and candidate boundary point j is selected in the previous stage i. This represents the width consistency constraint cost when selecting candidate boundary point k on this side in stage i+1. The weighting coefficient represents the cost of width consistency constraints.

[0111] For example, if the current side is the other side of the main road, then , , , , The values ​​were set to 0.35, 0.15, 0.1, 0.15, and 0.25 respectively.

[0112] Furthermore, the obstacle distance cost when selecting candidate boundary point k on this side in stage i+1. , represented as:

[0113] ,

[0114] In the formula, This represents the distance from the candidate boundary point k on this side of the current trajectory point i+1 to the nearest obstacle. This represents the distance scale parameter.

[0115] For example, distance scale parameter The value is 2 meters.

[0116] Furthermore, the trajectory following cost when selecting candidate boundary point k on this side in stage i+1. , represented as:

[0117] ,

[0118] In the formula, This represents the coordinates when k is selected as a candidate boundary point on this side in stage i+1. This represents the coordinates of the trajectory point in stage i+1. It represents Euclidean distance.

[0119] Furthermore, the distance cost between adjacent candidate boundary points when candidate boundary point k is selected in stage i+1 and candidate boundary point j was selected in the previous stage i is considered. , represented as:

[0120] ,

[0121] In the formula, This indicates the coordinates of point j selected as a candidate boundary point on that side in stage i.

[0122] Furthermore, the road width change cost when candidate boundary point k is selected in stage i+1 and candidate boundary point j is selected in the previous stage i is... , represented as:

[0123] ,

[0124] In the formula, This represents the coordinates of the previous trajectory point i-1.

[0125] Furthermore, the width consistency constraint cost when selecting candidate boundary point k on this side in stage i+1. , represented as:

[0126] ,

[0127] In the formula, This represents the coordinates of the optimal boundary point selected in stage i+1. This represents the coordinates of the candidate boundary point k selected on this side in stage i+1. Indicates the reference road width. The scale parameter represents the cost of width consistency constraints.

[0128] Preferably, the reference road width Represented as:

[0129] ,

[0130] In the formula, Indicates the historical average width. This represents the local road width in stage i; Indicates the mixed weighting coefficients, for example, =0.7. Among them,

[0131] Historical average width , represented as:

[0132] ,

[0133] In the formula, N represents the total number of trajectory points. This represents the coordinates of the initial passable area boundary point on one side of trajectory point n. This represents the coordinates of the initial passable area boundary point on the other side of trajectory point n.

[0134] Local road width at current trajectory point i , represented as:

[0135] ,

[0136] In the formula, w represents the radius of the sliding window; for example, w is set to 5.

[0137] For example, taking the current side as the main road side as an example, the specific process of obtaining the optimal boundary point sequence of the current side of the main road will be explained:

[0138] First, for initialization, in the final stage, the minimum cumulative cost of each candidate boundary point is its own cost; that is...

[0139] For stage i=N:

[0140] ,

[0141] ,

[0142] in, ;

[0143] In the formula, This represents the minimum cumulative cost to reach stage i and select candidate point j. The path represents the path with the minimum cumulative cost to reach stage i and select candidate point j; M represents the number of candidate boundary points, and N represents the number of trajectory points, i.e., the total number of stages.

[0144] Then, for each stage i, decrease from N-1 to 1:

[0145] ,

[0146] ,

[0147] Finally, determine the starting path with the minimum cumulative cost in the first phase, and backtrack forward. This yields the optimal boundary point sequence for the current side of the main road.

[0148] Preferably, width continuity inspection and local smoothing can also be performed based on the optimal boundary point sequence on both sides of the main road to eliminate abnormal boundary points.

[0149] For example, such as Figure 9 In the left part, the vehicle's driving trajectory points are black, green represents the boundary points of the passable area on the left side of the vehicle, and white represents the boundary points of the passable area on the right side of the vehicle. Figure 9 In the right-hand section, the black line represents the vehicle's trajectory, the green line represents the boundary of the passable area on the left, and the white line represents the boundary of the passable area on the right. It can be seen that the green and white passable area boundaries appropriately expand the road area, better reflecting real-world road conditions. When there are obstacles on the road, a dynamic programming algorithm is used to extract the passable area boundary points, more closely resembling the actual passable area boundaries and unaffected by road obstacles. For example, the green boundary in the lower left corner is unaffected by small obstacles and directly conforms to the curb.

[0150] S3. Based on the environmental map and the center line of the main road, obtain the optimal branch roads of the main road, and then obtain the passable area boundaries of the optimal branch roads of the main road.

[0151] During implementation, the optimal branch roads of the main road are obtained in the following way:

[0152] S31. Based on the environment map, obtain the main road topology path and identify the candidate branch topology paths of the main road.

[0153] Specifically, based on the topology map in the environmental map, the topological path of the main road can be extracted, and the topological paths of each candidate branch road can be identified. Each candidate branch road topological path is connected to the main road topological path; paths branching off from the main road topological path are called branch road topological paths, i.e., candidate branch road topological paths. In the topology map, the main road topological path obtained based on the dynamic programming algorithm is the path closest to the historical vehicle trajectory or the main road centerline. It should be noted that the candidate branch road topological path here is a topological edge.

[0154] Preferably, based on the topology paths and grid map of each candidate branch of the main road, candidate branches whose distance to the nearest obstacle is less than half the vehicle width are deleted.

[0155] For example, such as Figure 11 In the middle, line segments 2-21, 4-22, 6-23, 7-24, and 9-25 are not passable because the distance from forks 5-26 and 8-27 to the nearest obstacle is less than half the width of the vehicle. Therefore, they are deleted.

[0156] S32. Based on the topological paths of each candidate branch of the main road, obtain the topological paths of each passable candidate branch of the main road.

[0157] In practice, the topology paths of each passable candidate branch on the main road are obtained in the following way:

[0158] S321. Based on the environment map, obtain all topological path endpoints, and then obtain the passable endpoints; where an endpoint is the endpoint of only one connecting path;

[0159] Specifically, the drivability of the endpoints of each topological path is determined in the following way:

[0160] Based on the path points on the current topological path endpoint, construct the covariance matrix of the X and Y coordinates;

[0161] Based on the covariance matrix, the first and second eigenvalues ​​are obtained, and then the two eigenvectors corresponding to the larger eigenvalue are taken as the main direction of the path.

[0162] Construct the equation of the straight line based on the main direction of the path and the coordinates of the endpoint;

[0163] Based on the equation of a straight line, a line segment of a set length threshold is extended from the end point along the tangent direction. If the line segment does not pass through any obstacles or the distance to the nearest obstacle is greater than half the width of the vehicle, the end point of the current topological path is determined to be passable; otherwise, the end point of the current topological path is determined to be impassable.

[0164] For example, such as Figure 11 As shown, path points numbered 11, 15, 16, 17, 18, 19, and 20 are passable end points.

[0165] S322. Based on each passable end point, obtain the shortest path from each candidate branch topology path of the main road to the exit, and take each candidate branch topology path of the main road and its shortest path to the exit as each passable candidate branch topology path.

[0166] Specifically, based on Dijkstra's shortest path method, the shortest path from each candidate branch topology path of the main road to each passable end point is obtained, and then the shortest path from each candidate branch topology path of the main road to the exit is obtained.

[0167] Among them, the shortest path from the candidate branch topology path to the shortest length is taken as the shortest path from the candidate branch topology path to the exit.

[0168] In Dijkstra's shortest path method, all endpoints of the topological path are used as nodes, the connecting paths between adjacent nodes are used as edges, and the pixels on each topological path are used as weights to construct a topological graph. During path search, the end point of the branch path is used as the starting point, and all passable end points are used as the ending points.

[0169] Preferably, the shortest path containing all or part of the main road topology path is deleted.

[0170] For example, such as Figure 11 As shown, the known possible branch road segments are 2-21, 4-22, 6-23, 7-24, and 9-25; record the endpoints connecting all possible branch road segments to the main road (e.g., ...). Figure 11 (Points numbered 2, 4, 6, 7, 9), then using the end point of each branch line segment (e.g., Figure 11 Starting from points numbered 21, 22, 23, 24, and 25, each passable endpoint (e.g., ...) Figure 11 Using points numbered 11, 15, 16, 17, 18, 19, and 20 as endpoints, the number of pixels in each topological road segment is used as the weight; the smaller the number of pixels, the smaller the weight of each path segment. Dijkstra's algorithm from graph theory is used to find the shortest topological path from the end point of each branch road segment to the accessible end point. If the shortest topological path contains a main road topological path, it does not conform to the intuitive definition of a branch road and will be deleted. Figure 11 As shown, among all the paths to the end of the line segments, the shortest path is from the fork in the road labeled 2 to the end of the line segment labeled 15; the shortest path is from the fork in the road labeled 4 to the end of the line segment labeled 11; and the shortest path is from the fork in the road labeled 9 to the end of the line segment labeled 19.

[0171] S33. Based on the topology and environmental map of each passable candidate branch of the main road, obtain the passable area boundary of each passable candidate branch.

[0172] It should be noted that the passable area boundaries of each passable candidate branch are determined according to the method for determining the passable area boundary of the main road in step S2; wherein, the topological path of each passable candidate branch is used as the centerline of each passable candidate branch.

[0173] S34. Based on the topological paths of each passable candidate branch road of the main road and the boundaries of its passable area, construct a multi-objective optimization function and constraints, solve for them, and obtain the optimal branch roads of the main road.

[0174] In practice, the multi-objective optimization function includes minimizing the rate of change of curvature, minimizing the path length, and maximizing the road width; the constraints include collision avoidance constraints and turning radius constraints.

[0175] Specifically, the objective function for minimizing the rate of change of curvature is expressed as:

[0176] ,

[0177] In the formula, The objective function representing the rate of change of curvature is... Let Q represent the q-th path point and the rate of curvature change of the passable candidate branch topology path, where Q represents the number of path points selected on the passable candidate branch topology path.

[0178] Understandably, the smaller the rate of curvature change of the passable candidate branch topology path, the smoother the vehicle turns and the higher the ride comfort.

[0179] Specifically, the objective function for minimizing path length is expressed as:

[0180] ,

[0181] In the formula, Represents the objective function for all path lengths. represents the length of the m-th topological segment in the shortest path from the passable candidate branch topological path to the exit, and M represents the total number of topological segments in the shortest path from the passable candidate branch topological path to the exit.

[0182] It is understandable that the shorter the distance from the topological path to the exit (any passable end point) of the topological map, the better the topological path is.

[0183] For example, such as Figure 10 In the topology diagram shown, orange ABCD is the main road topology segment, purple BF and CE are candidate branch roads, and G is the exit point of the topology map. The topology route from branch road segment CE to exit point G is EF-FG, and the topology route from branch road segment BF to exit point is FG. Obviously, branch road BF is closer to exit point F, so branch road BF is the preferred choice.

[0184] Specifically, the objective function for maximizing road width is expressed as:

[0185] ,

[0186] In the formula, The objective function representing the road width is... Indicates the number of passable candidate branch paths in the topology. The average road width of the segment topology line segment, This represents the total number of topological segments in the passable candidate branch road topological path. The average road width of the passable candidate branch road topological path is calculated based on the initial passable area boundary.

[0187] Understandably, the wider the road at a fork in the road, the lower the probability of a vehicle colliding with an obstacle, and the safer the vehicle is to travel on that path.

[0188] For example, such as Figure 10 In the diagram, the topological path from branch point C to exit point G is CEFG. Obviously, its topological road width is smaller than that of the topological path from branch point B to exit point G, which is BFG. Therefore, branch point BF is preferred.

[0189] Specifically, the collision avoidance constraint is expressed as:

[0190] ,

[0191] In the formula, d represents the radius of the vehicle's coverage circle, and d represents the distance of the vehicle to the nearest obstacle.

[0192] Understandably, collision avoidance constraints require the radius of the covered circle to be greater than the distance to the nearest obstacle to ensure that the vehicle does not collide with the obstacle.

[0193] Specifically, the turning radius constraint is expressed as:

[0194] ,

[0195] In the formula, This represents the radius of the inscribed circle of the rectangular area of ​​the road. This indicates the vehicle's minimum turning radius.

[0196] Understandably, the turning radius constraint is such that the radius of the inscribed circle of the road rectangle region extended by the topological path is greater than the vehicle's minimum turning radius, ensuring that the vehicle can turn smoothly.

[0197] Specifically, during the solution process, weights are assigned based on the importance of each objective function in the optimization problem, and then a linear combination is performed to seek the optimal solution. This yields a comprehensive score for each candidate branch, and the candidate branch with the highest comprehensive score and a set number of such scores is selected as the optimal branch. In this embodiment, the data in the objective optimization function is standardized by setting extremely large and extremely small indices. The rate of change of curvature and path length are extremely small indices, while the road width is an extremely large indice.

[0198] More specifically, this embodiment employs a linear weighting method, assigning weights to each objective function based on their importance in the optimization problem, to obtain the comprehensive scoring function F for the candidate paths:

[0199] ,

[0200] in, ;

[0201] In the formula, , , The objective functions are respectively , , The standardized objective function, , , The objective functions are respectively , , The weighting coefficients.

[0202] The objective function is standardized in the following way:

[0203] For extremely large metrics:

[0204] ,

[0205] For extremely small metrics:

[0206] ,

[0207] In the formula, , These represent the original indicator and the standardized indicator, respectively. , These represent the maximum and minimum values ​​of this index among all candidate branch paths, respectively.

[0208] Specifically, after obtaining the comprehensive score for each branch, the candidate branch with the highest score among a set number of branches is selected as the optimal branch.

[0209] Preferably, each optimal branch road boundary will have two intersection points with the main road boundary. The sequential positions of these two intersection points on the main road boundary are calculated, and the boundary points between these two corresponding sequential positions are deleted to open the corresponding branch road gaps. Based on the corresponding intersection points, the corresponding sequential positions of the left and right branch road boundary points are found, and the branch road boundary points between the initial position of the branch road boundary point and the corresponding index of the intersection point are deleted, so that the main road boundary and the branch road boundary can be smoothly connected. The final result is as follows: Figure 12 As shown.

[0210] For example, such as Figure 12 Because branch road 6-23 has no passable endpoint connected to it, it is impassable; branches road 2-21 and 4-22 meet the requirements for local branch roads; branch road 9-25 is better than 7-24 and better meets the requirements for local branch roads, generating a passable area boundary as follows. Figure 13 As shown.

[0211] Preferably, each optimal branch road is connected to the main road. If a branch road intersects with the main road in the environment map, the intersection point (e.g., ...) is connected to the main road. Figure 14 The point labeled 6 is converted into a global pose, and its index in the vehicle trajectory is recorded as the connection point between the main road and the side road. The portion outside the intersection of the side roads (such as...) is also converted into a global pose. Figure 14 The red line segment between point 6 and point 23 is also converted into a global pose. If the branch does not intersect with the main path, it is necessary to calculate the shortest distance between the position of the zero pixel on the main path and the position of the non-zero pixel, and also calculate the position (i.e., the index value) of the nearest non-zero pixel to that pixel. At this time, it is necessary to record the branch points on the side closer to the main path (such as...). Figure 14 The nearest non-zero pixel index value (e.g., the intersection of the red and light green line segments, labeled 2) Figure 14 The intersection of the red and black line segments, labeled 40, is converted into global coordinates, its index in the vehicle trajectory points is cached, and the entire branch road (such as...) is... Figure 14 The local coordinates of the red line segment between point 2 and point 21 are converted into global pose, and the connection relationship with its nearest trajectory is recorded. To ensure that the selected branch road is convenient for vehicles to pass, the cosine theorem is used to make the branch road points far away from the main road (such as...)... Figure 14 The point number is 21) and the fork points near the main road (such as Figure 14 The point number is 2) and the nearest vehicle trajectory point (e.g. Figure 14The points are labeled 40 (maintaining an obtuse angle). Because each branch road corresponds to a unique vehicle trajectory index, whenever the branch road corresponding to a vehicle trajectory point changes, the corresponding branch road information will be updated. The above grid map contains 5 branches road, such as... Figure 14 The red line segments between point 6 and point 23, and between point 2 and point 21, and between point 41 (42,43) and point 4 (7,9) (the connection principle is similar to that of red number 40-2), the black line segments are the main road centerline, and the light green line segments are the main road topology path.

[0212] S4. Under the geometric and kinematic constraints of the vehicle, based on the traversable area boundaries of the main road and its optimal branch roads, the traversable optimal branch roads of the main road are obtained, and then the traversable area boundaries of the main road and its optimal branch roads are obtained.

[0213] During implementation, the optimal passable branch roads of the main road are obtained in the following way:

[0214] S41. Based on the optimal branch roads of the main road and their passable area boundaries, perform passability verification for each optimal branch road, including:

[0215] S411. Perform geometric smoothing on the current optimal branch to obtain the initial reference path of the current optimal branch; wherein, the reference path includes each path point and its geometric information;

[0216] S412. Based on the initial reference path of the current optimal branch, construct vehicle kinematic constraints, and based on the corresponding passable area boundary, construct vehicle collision constraints; wherein, vehicle kinematic constraints include vehicle kinematic equation constraints and vehicle control quantity constraints.

[0217] S413. Based on vehicle kinematic constraints and vehicle collision constraints, construct a path planning optimization model with the objective function of minimizing vehicle control cost, control smoothing cost, path cost and safe centering cost, and solve it to obtain the current optimal branch's drivability state.

[0218] S42. Based on the passability status of each optimal branch, obtain each passable optimal branch of the main road.

[0219] In specific implementation, in step S411, the initial reference path of the current optimal branch is obtained in the following way:

[0220] Discretize the current optimal branch to obtain each path point and its normal vector of the current optimal branch;

[0221] Using the lateral offset of each path point relative to the current optimal branch as the optimization variable, a geometric optimization model is constructed with the objective function of minimizing the lateral deviation cost, heading change cost, and curvature change cost, and then solved to obtain the optimal lateral offset of each path point.

[0222] Based on the optimal lateral offset of each path point and the normal vector of the current optimal branch, an initial reference path is obtained, and then the geometric information of each path point on the initial reference path is obtained, including the curvature and heading angle of each path point.

[0223] Specifically, the geometric optimization model is expressed as:

[0224] ,

[0225] in, ;

[0226] In the formula, Indicates the first The lateral offset of each path point relative to the current optimal branch. Indicates the first The change in heading angle between each path point and its adjacent path points Indicates the first +1 change in heading angle between the path point and its adjacent path points Indicates the first +1 lateral offsets of path points relative to the current optimal branch, where ds represents the arc length between adjacent path points. , , These represent the weighting coefficients for lateral deviation cost, heading change cost, and curvature change cost, respectively.

[0227] For example, the weighting coefficients for lateral deviation cost, heading change cost, and curvature change cost are set to 0.3, 0.4, and 0.3, respectively.

[0228] Specifically, the lpopt solver is used to solve the geometric optimization model of unconstrained nonlinear optimization.

[0229] Specifically, the vehicle control quantity is the steering wheel angle.

[0230] In practical implementation, the path planning optimization model is represented as follows:

[0231] ;

[0232] In the formula, , These represent the initial reference path number 1 and 2 respectively. , +1 waypoint vehicle steering wheel angle Indicates the initial reference path number. Lateral deviation of each path point Indicates the initial reference path number. The lateral position of the rear axle center of the vehicle at each path point , These represent the initial reference path number. The right and left boundary points of the passable area corresponding to each path point This indicates the number of path points in the initial reference path; , These represent the initial reference path number. , +1 state variables for path points , , These represent the initial reference path number. The state transition matrix, control input matrix, and constant term vector for each path point. Indicates the initial reference path number. The lateral position vector of the vehicle envelope center at each path point. , Indicates the initial reference path number. Minimum and maximum safe boundaries of the vehicle envelope circle for each path point. , This indicates the minimum and maximum values ​​of the vehicle's steering wheel angle. , These represent the starting and ending points of the path, respectively. , , , These represent the weighting coefficients for vehicle control cost, control smoothing cost, path cost, and safety centering cost, respectively.

[0233] Specifically, vehicle kinematic constraints are constructed in the following manner:

[0234] The vehicle state x in the Frenet coordinate system is represented by heading deviation and lateral deviation as follows:

[0235] ;

[0236] In the formula, This indicates the angle of deviation between the vehicle's heading and the tangent direction of the reference path. This represents the lateral distance from the vehicle's center of gravity to the reference path.

[0237] The control input u is the steering wheel angle. The state equation is The derivative with respect to time:

[0238] ;

[0239] In the formula, v represents the longitudinal velocity of the vehicle. This refers to the vehicle's wheelbase. This represents the angular velocity of the reference path.

[0240] From Frenet coordinate system ,in, Given the curvature along the reference path, the derivative of the state with respect to the reference path length s can be obtained:

[0241] ;

[0242] Expanding the above equation with a first-order Taylor series at the reference path, and linearizing it, we get:

[0243] ;

[0244] In the formula, Indicates the steering wheel angle of the reference path. This indicates the reference control value.

[0245] Discretization using the Euler method, step size :

[0246] ;

[0247] Where I represents the identity matrix.

[0248] Notation:

[0249] ;

[0250] in, , , These represent the initial reference path number. The state transition matrix, control input matrix, and constant term vector for each path point.

[0251] At the same time, the control quantity must also meet the constraints of the rotation range:

[0252] ;

[0253] Linearized vehicle kinematic constraints are obtained.

[0254] Specifically, vehicle collision constraints are constructed in the following manner:

[0255] like Figure 15As shown, the vehicle's shape is represented by three circles covering it. The arrangement of the three circles is as follows: the rear axle circle is located at the center of the rear axle, the center of mass circle is located at the center of mass, and the front axle circle is located at the center of the front axle. At each discrete point, the offset of the three circle centers from the reference position is calculated, and upper and lower bounds are set according to obstacles.

[0256] ;

[0257] In the formula, , , These represent the distances from the center of the vehicle's rear axle to the centers of the three circles. , , These are the reference paths. The lateral positions of the rear axis center, centroid center, and front axis center of each path point; , Indicates the reference path number The minimum and maximum safe boundaries of the vehicle envelope circle for each path point are obtained by using the traversable area boundary, the radius of the envelope circle, and the safety margin.

[0258] Specifically, the path planning optimization model can be solved as a quadratic programming problem using the Ipopt nonlinear solver, which is a small-scale and fast process.

[0259] In practice, if the path planning optimization model of the optimal branch has no solution, the current optimal branch is inaccessible; otherwise, the current optimal branch is accessible.

[0260] Preferably, for impassable side roads, the boundary opening of the side road is deleted, and two boundary points that should have been open are connected on the boundary of the main road to ensure that the closed boundary is smooth and continuous.

[0261] For example, the originally extracted passable area boundary contained three branching paths, while... Figure 16 In the middle, passing through fork 2-21, vehicles need to make a sharp turn. Using the aforementioned vehicle local planning method, it is impossible to generate a collision-free smooth path that satisfies vehicle kinematic constraints. Therefore, the traversable area boundary of this fork is closed. Figure 15 In the diagram, light green paths 50-22 and 51-25 represent the local planning paths for vehicles passing through forks 50-22 and 51-25, respectively. By conducting a detailed analysis of the vehicle accessibility of these forks, the topology is simplified, generating a more accurate boundary model of the passable area.

[0262] Compared with existing technologies, this embodiment provides a method for extracting the drivable area boundary based on an environmental map. It acquires an environmental map integrating a topological map and a raster map, smooths historical vehicle trajectories to obtain the main road centerline, and then obtains the initial drivable area boundary of the main road. A dynamic programming algorithm is then used to obtain the drivable area boundary of the main road. Next, based on the environmental map and the main road centerline, the optimal branches of the main road are obtained, and the drivable area boundaries of each optimal branch are obtained. Finally, under the geometric and kinematic constraints of the vehicles, based on the drivable area boundaries of the main road and its optimal branches, the following method is obtained: The optimal passable branch roads of the main road are identified, and the passable area boundaries of the main road and its optimal passable branch roads are obtained. By utilizing the continuity and width distribution of the passable area boundaries, the continuity and shape of the road can be effectively reflected. Modeling the passable area boundary model helps improve the accuracy of passable area boundary extraction, making it more consistent with the real road environment. It can be better applied to outdoor environmental scenarios and is also suitable for off-road environments with complex obstacles and irregular shapes. It can reduce the computation time for passable area boundary extraction when unmanned vehicles are autonomously navigating, and help improve the smoothness and safety of unmanned vehicle driving.

[0263] Example 2

[0264] To illustrate the effectiveness and correctness of Example 1, a specific embodiment is provided for explanation. Experimental Platform Introduction: The test platform based on the ground-based unmanned aerial vehicle (UAV) mobility system is used to collect relevant data from the actual operating environment of the ground-based UAV platform, providing raw data for experimental testing. This platform consists of a control system, an autonomous navigation system, and an action system. The autonomous navigation system is integrated into the action system, while the control system is relatively independent, allowing operators to interact with the test platform using the control system. The control system consists of a remote control unit and a local control unit. The remote control unit mainly consists of a ruggedized computer and wireless network communication equipment, using control software deployed on the computer to achieve platform status monitoring, remote driving, and autonomous following functions. The mobility system uses hybrid power and pure electric drive, with a high-density lithium battery as the power source. The autonomous navigation system is the core of the test platform and mainly consists of sensing sensors, an industrial control computer, and a controller. The sensing sensors mainly include millimeter-wave radar, cameras, lidar, inertial navigation systems, and GPS receivers, primarily to obtain positioning information of the surrounding environment and the vehicle. The main function of an industrial control computer is to process environmental data, build maps and plan vehicle routes, and finally generate control commands and send them to the vehicle controller and actuators.

[0265] Extraction of traversable area boundaries and verification of branch road traversability tests in off-road environments

[0266] exist Figure 17In the middle, the left part shows the results of extracting the boundaries of the passable area considering vehicle passability and smoothness, while the light green area in the right part is the main road that vehicles can pass (artificially labeled).

[0267] exist Figure 17 In the left image, the result of extracting the main road passable area boundary using the method in Example 1 is shown. In the right image, the white area represents the detected main road area enclosed by the passable area boundary. To more intuitively demonstrate the accuracy of the passable area boundary extraction in Example 1, it can be viewed as a semantic segmentation problem. In semantic segmentation, the Intersection over Union (IOU) between the ground truth label and the predicted result is a standard accuracy metric. Example 1 extracted the passable area boundaries for ten off-road intersection scenarios and calculated an IOU of 88.8%, indicating that the passable area boundary extraction algorithm of Example 1 has high accuracy.

[0268] exist Figure 18 The green and white lines represent the boundaries of the main road's passable area, the black lines represent the vehicle's trajectory, the red lines represent the extracted branch road paths, and the light green lines represent the vehicle's local planned path, indicating that vehicles can pass through this branch road. The branch road where the orange circle is located is one where vehicles cannot pass due to vehicle size and kinematic constraints. The purple and yellow lines represent the branch road boundaries.

[0269] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0270] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for extracting the boundaries of traversable areas based on an environmental map, characterized in that, Includes the following steps: Obtain an environment map that integrates topology and raster maps; Based on the environmental map, historical vehicle trajectories are smoothed to obtain the centerline of the main road, thereby obtaining the initial passable area boundary of the main road. Then, a dynamic programming algorithm is used to obtain the passable area boundary of the main road. The passable area boundary includes the sequence of boundary points on both the left and right sides of the passable area. The process of obtaining the passable area boundary of the main road using the dynamic programming algorithm includes the following steps: Based on the main road centerline, its initial passable area boundary, and a set width interval, candidate boundary point strips are generated on the left and right sides; wherein, each candidate boundary point strip represents a candidate boundary point on the side corresponding to a trajectory point; specifically, for each trajectory point on the main road centerline, the left and right boundary points of the passable area corresponding to each trajectory point are generated along the normal direction of the corresponding trajectory point with a set width interval, and a set number of candidate boundary points are generated to form the left and right candidate boundary point strips. Based on the main road centerline and the candidate boundary point zone on one side, the optimal boundary point sequence on one side of the main road is obtained through dynamic programming optimization of one side boundary. Based on the optimal boundary point sequence on one side of the main road, the centerline of the main road and the candidate boundary point zone on the other side, the optimal boundary point sequence on the other side of the main road is obtained through dynamic programming optimization of the boundary on the other side. Based on the optimal boundary point sequence on both sides of the main road, the traversable area boundary of the main road is obtained. In dynamic programming optimization, if the current side is on the main road, the single-step cost function for that side includes obstacle distance cost, boundary smoothness cost, and trajectory following cost; if the current side is on the other side of the main road, the single-step cost function for that side includes obstacle distance cost, boundary smoothness cost, trajectory following cost, and width consistency constraint cost; where the width consistency constraint cost is expressed as: ， In the formula, Indicates the stage i+ 1. Select as the candidate boundary point on this side. k The cost of width consistency constraints during time Indicates the stage i+ 1. Select the optimal boundary point coordinates on one side. Indicates the stage i+ 1. Selected candidate boundary points on this side k coordinates Indicates the reference road width. The scale parameter representing the cost of width consistency constraints. Representation phase i+ The coordinates of the trajectory point 1; where, Reference road width Represented as: ， In the formula, Indicates the historical average width. Indicates the stage i The width of the local road; Represents the mixed weighting coefficients; where, Historical average width , is represented as: ， In the formula, N represents the total number of trajectory points. Represents trajectory points n The coordinates of the initial passable area boundary point on one side. Represents trajectory points n The coordinates of the initial passable area boundary point on the other side; In the stage i Local road width , is represented as: ， In the formula, w Indicates the radius of the sliding window; Based on the environmental map and the center line of the main road, the optimal branch roads of the main road are obtained, and then the passable area boundaries of the optimal branch roads of the main road are obtained. Under the geometric and kinematic constraints of the vehicle, based on the traversable area boundaries of the main road and its optimal branch roads, the traversable optimal branch roads of the main road are obtained, and thus the traversable area boundaries of the main road and its optimal branch roads are obtained.

2. The method for extracting the boundary of a traversable area based on an environmental map according to claim 1, characterized in that, The initial passable area boundary of the main road is obtained in the following way: In the environmental map, the historical vehicle trajectory points are smoothed by B-spline curve fitting to obtain the center line of the main road; Based on the centerline of the main road, the normal vector of each trajectory point is obtained; based on the environmental map, the distance from each trajectory point to the nearest obstacle is obtained, and then the road width on the left and right sides of each trajectory point is obtained; Each trajectory point on the center line of the main road is extended along its normal direction and based on the corresponding width of the left and right roads to obtain the initial passable area boundary of the main road.

3. The method for extracting the boundary of traversable areas based on environmental maps according to claim 1, characterized in that, The optimal boundary point sequences on both sides of the main road are obtained using the following methods: Based on the main road centerline and the candidate boundary point zone on its current side, a probabilistic graphical model and a single-step cost function are constructed for the current side. The probabilistic graphical model is constructed based on each trajectory point on the main road centerline to build each stage. The state of each stage is the candidate boundary point of the corresponding trajectory point on the current side, and the state value is the minimum cumulative cost. The minimum cumulative cost is the minimum cumulative cost to reach the current stage and select the current candidate node, which is obtained based on the single-step cost function. Based on the constructed current-side probabilistic graphical model and single-step cost function, dynamic programming with inverse recursion is used to obtain the optimal boundary point sequence of the current side of the main road.

4. The method for extracting the boundary of traversable areas based on environmental maps according to claim 1, characterized in that, The optimal branches of the main road are obtained in the following way: Based on the environmental map, the main road topology path is obtained, and the candidate branch topology paths of the main road are identified. Based on the topological paths of each candidate branch of the main road, the topological paths of each passable candidate branch of the main road are obtained. Based on the topological paths and environmental maps of each passable candidate branch of the main road, the passable area boundaries of each passable candidate branch are obtained. Based on the topological paths of each passable candidate branch of the main road and the boundaries of its passable areas, a multi-objective optimization function and constraints are constructed and solved to obtain the optimal branch of the main road.

5. The method for extracting the boundary of a traversable area based on an environmental map according to claim 4, characterized in that, The multi-objective optimization function includes minimizing the rate of change of curvature, minimizing the path length, and maximizing the road width; the constraints include collision avoidance constraints and turning radius constraints.

6. The method for extracting the boundary of a traversable area based on an environmental map according to claim 1, characterized in that, The optimal passable branch roads of the main road are obtained in the following way: Based on the optimal branch roads of the main road and their passable area boundaries, the passability of each optimal branch road is verified, including: Geometrically smooth the current optimal branch to obtain the initial reference path of the current optimal branch; wherein, the reference path includes each path point and its geometric information; Based on the initial reference path of the current optimal branch, vehicle kinematic constraints are constructed, and vehicle collision constraints are constructed based on the corresponding passable area boundaries; among them, vehicle kinematic constraints include vehicle kinematic equation constraints and vehicle control quantity constraints. Based on vehicle kinematics constraints and vehicle collision constraints, a path planning optimization model is constructed with the objective function of minimizing vehicle control cost, control smoothing cost, path cost and safe centering cost, and then solved to obtain the current optimal branch's drivability state. Based on the passability status of each optimal branch, the passable optimal branches of the main road are obtained.

7. The method for extracting the boundary of a traversable area based on an environmental map according to claim 6, characterized in that, The initial reference path for the current optimal branch is obtained in the following way: Discretize the current optimal branch to obtain each path point and its normal vector of the current optimal branch; Using the lateral offset of each path point relative to the current optimal branch as the optimization variable, a geometric optimization model is constructed with the objective function of minimizing the lateral deviation cost, heading change cost, and curvature change cost, and then solved to obtain the optimal lateral offset of each path point. Based on the optimal lateral offset of each path point and the normal vector of the current optimal branch, an initial reference path is obtained, and then the geometric information of each path point on the initial reference path is obtained, including the curvature and heading angle of each path point.

8. The method for extracting the boundary of traversable areas based on environmental maps according to claim 2, characterized in that, The trajectory points were smoothed by using cubic clamping B-spline curve fitting.

Images