A local path planning method based on a segmentation map evaluation function

Abstract

The application discloses a local path planning method based on a segmentation map evaluation function, and comprises the following steps: using a depth camera and a laser radar of an unmanned vehicle mobile platform, and using an external computer to obtain a semantic segmentation result of a non-paved road in the field by using an HRNet+OCR semantic segmentation network; performing inverse perspective transformation on the semantic segmentation result to obtain a bird's eye view and performing binary processing to calculate a corresponding position of the unmanned vehicle in the bird's eye view; calculating a total number of pixels of a preselected path as a first evaluation function, and scoring; introducing a path point direction of an optimal path as a second evaluation function, combining the first evaluation function to obtain an overall score, and selecting an optimal path to move. The method can effectively complete local path planning and obstacle avoidance operation of the unmanned vehicle in the field when a front part or the whole front part is an impassable or obstacle region.

Description

Technical Field

[0001] This invention relates to the field of machine vision and autonomous driving image processing, and in particular to a local path planning method based on a segmentation graph evaluation function. Background Technology

[0002] When autonomous vehicles are driving on unstructured roads in the wild, if the area ahead is partially or entirely impassable or contains obstacles, local path planning is required. If the mobile platform is traveling on surfaces such as mud, cement, or sand, traditional local path planning algorithms such as DWA can be used to determine the cost function of the predicted path and optimize it using a grid map and the corresponding cost map, thereby achieving dynamic obstacle avoidance. However, in situations involving grass or shrubs, the LiDAR may malfunction due to obstruction by weeds and shrubs, preventing the correct generation of grid maps and corresponding cost maps, thus hindering proper local path planning and dynamic obstacle avoidance. Summary of the Invention

[0003] The purpose of this invention is to provide a local path planning method based on a segmentation map evaluation function, in order to solve the problem that in the existing technology, traditional local path planning algorithms may fail to function properly in the case of grassland, shrubs, etc., due to the occlusion of weeds, shrubs, etc., causing the LiDAR to malfunction, thus failing to generate the grid map and the corresponding cost map correctly, and thus failing to perform local path planning and dynamic obstacle avoidance correctly.

[0004] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides a local path planning method based on a segmentation graph evaluation function, comprising the following steps:

[0005] Video images are acquired using the depth camera and LiDAR of the unmanned vehicle mobile platform, and semantic segmentation results of unpaved roads in the field are obtained using the HRNet+OCR semantic segmentation network.

[0006] The semantic segmentation results are transformed into a bird's-eye view through inverse perspective transformation and then binarized to calculate the corresponding position of the unmanned vehicle in the bird's-eye view.

[0007] The total number of pixels traversed by the pre-selected path is calculated and used as the first evaluation function for scoring;

[0008] The path point direction is introduced as a second evaluation function for the preferred path. Combined with the first evaluation function, the overall score is obtained, and the optimal path is selected for movement.

[0009] In a second aspect, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method described in the first aspect.

[0010] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0011] Fourthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0012] Compared with the prior art, the beneficial effects of the present invention are: the local path planning algorithm based on the segmentation map evaluation function of the present invention adopts the method of optimizing the proportion of pre-selected path pixels in the binarized bird's-eye view using the evaluation function, which effectively solves the problem that the traditional local path planning algorithm in the prior art may fail to generate grid maps and corresponding cost maps correctly and fail to perform local path planning and dynamic obstacle avoidance correctly in the case of grassland, shrubs and other situations where the lidar cannot work properly due to the occlusion of obstacles such as weeds and shrubs. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the overall process of the present invention.

[0014] Figure 2 This is a schematic diagram of the local path planning method using the segmentation graph evaluation function.

[0015] Figure 3 This is a schematic diagram of the pre-selected path judgment and optimization process. Detailed Implementation

[0016] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0017] like Figure 1 As shown, a local path planning method based on a segmentation graph evaluation function according to the present invention includes the following steps:

[0018] Step 1: Use the depth camera and LiDAR of the unmanned vehicle mobile platform to acquire video images. The depth camera is installed in the center of the front of the unmanned vehicle. Use an external computer to obtain the semantic segmentation results of the unstructured road in the wild using the HRNet+OCR semantic segmentation network.

[0019] Step 2: The semantic segmentation result is transformed into a bird's-eye view through inverse perspective transformation and then binarized. The corresponding position of the autonomous vehicle in the bird's-eye view is calculated. Specifically:

[0020] The semantic segmentation results of unstructured roads in the field are subjected to inverse perspective transformation to convert them into a bird's-eye view. Then, binarization is performed. In the binarized image, the selection of the four image points and four object points is symmetrical about the center line of the camera's field of view. Therefore, for the bird's-eye view result, its bottom midpoint can be regarded as the current position of the depth camera on the moving platform, thus determining the corresponding position of the unmanned vehicle in the bird's-eye view. The center line of the bird's-eye view can be regarded as the direction directly in front of the moving platform.

[0021] Step 3: Calculate the total number of pixels traversed by the pre-selected path and use this as the first evaluation function for scoring; specifically:

[0022] The binarized image is listed with multiple possible paths, such as... Figure 2 As shown, this method sets the reference trajectory of the mobile platform as a directed line segment, determines the angle of the mobile platform's reference trajectory, and calculates the reference trajectory angle at each control command interval T. s The mobile platform is steered and moved within the system. The proportion of pixels traversed by the pre-selected path relative to the total pixels of all paths is used as the first evaluation function for optimization. Specifically, the number of pixels in the passable area is used as the metric. A ray is emitted from the location of the camera in the binarized bird's-eye view; the direction traversed by the ray corresponds to a less obstructed, flatter, and more open terrain suitable for passage. The mobile platform can rotate in this direction to move away from areas that are impassable or blocked. Specifically, starting from the midpoint of the baseline in the binarized bird's-eye view, n line segments P1 to P2 are drawn, with an interval of ψ and a radius of R pixels, encompassing both sides of the midline. n The scoring function for each line segment is:

[0023]

[0024] Among them, Pixel (P i () represents the total number of pixels traversed by the line segment.

[0025] Step 4: Introduce the path point direction as the second evaluation function for the preferred path, and combine it with the first evaluation function to obtain the overall score and select the optimal path to move. This method can effectively complete local path planning and obstacle avoidance operations for unmanned vehicles in the wild when the area in front is partly or entirely impassable or has obstacles.

[0026] like Figure 3As shown, if none of the n line segments in the current bird's-eye view touches any impassable areas or obstacles (i.e., each line segment has the same score evaluation function), the moving platform continues to move forward. When a line segment detects an impassable area or obstacle, the line segment contracts inward along the center to the edge of the obstacle. At this point, the direction of the longest line segment is determined, which is the direction the moving platform needs to rotate and move in. If there are multiple line segments of the same length, the line segments are extended outward until they reach an impassable area, obstacle, or the boundary of the detected image. At this point, the direction pointed to by the longest line segment (i.e., the line segment with the largest score evaluation function) is selected as the direction the moving platform needs to turn in. If, after extension, there are still several line segments of the same length, such as... Figure 3 As shown, another evaluation criterion will be introduced, namely, a path point direction score function will be introduced as a second evaluation function for the preferred path. Combined with the first evaluation function, the overall evaluation function will be obtained. The evaluation function for the path direction score is as follows:

[0027]

[0028] Where ε is a local minimum, ψ GPS ψ represents the direction of the next path point. i Let α and β represent the direction the current mobile platform is facing, and introduce weighting coefficients α and β. The overall score evaluation function is as follows:

[0029] Score i =αScore Path,i +βScore Dir,i

[0030] At this point, the optimal path is selected based on the highest score, and the mobile platform moves in this direction, thus ending the local path planning.

[0031] This embodiment verifies the design algorithm using the RIA-R100 mobile platform manufactured by Shanghai Silicon Step Scientific Instruments Co., Ltd. The RIA-R100 mobile platform is equipped with a Velodyne VLP-16 LiDAR, an Intel RealSense D435 depth camera, a U-Blox 7 GPS module, an HC-SR04 ultrasonic radar, and a SLAMTEC RPLIDAR-A2 LiDAR. The connected laptop uses a 2.60GHz Intel i7-9750H processor, 16GB of RAM, an NVIDIA RTX 2060 graphics card, and runs Windows 10.

[0032] This invention employs an evaluation function to optimize the proportion of pre-selected path pixels in the binarized bird's-eye view. This effectively solves the problem in existing technologies where traditional local path planning algorithms may fail to function properly in situations such as grasslands and shrubs due to the obstruction of obstacles like weeds and shrubs, resulting in the inability to generate grid maps and corresponding cost maps correctly, and thus failing to perform dynamic obstacle avoidance for local path planning.

[0033] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several modifications and improvements without departing from the inventive concept of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. A local path planning method based on a segmentation map evaluation function, characterized in that, The method comprises the following steps: A depth camera and a laser radar of the unmanned vehicle mobile platform are used to acquire video images, and a HRNet+OCR semantic segmentation network is used to acquire a semantic segmentation result of a non-paved road in the wild; The semantic segmentation result is subjected to inverse perspective transformation to an aerial view and is subjected to binary processing, and a corresponding position of the unmanned vehicle in the aerial view is calculated; A total number of pixels of a preselected path is calculated as a first evaluation function, and scoring is performed, specifically: The possible paths of the binarized image are enumerated, the reference trajectory of the mobile platform is set as a directed line segment, the angle of the reference trajectory of the mobile platform is judged, and the steering and movement of the mobile platform are performed in each control command interval T s The total number of pixels of the preselected path is calculated, the ratio of the total number of pixels of the preselected path to the total number of pixels of all paths is taken as a first evaluation function, and the preselection is performed, that is, the number of pixels of the passable area is taken as a measurement standard, a ray is emitted from the position of the bird's-eye view camera after binarization, and the more pixels the ray passes through, the direction is turned to by the mobile platform. With the midpoint of the bottom line of the bird's-eye view after binarization as the starting point, containing the midline on both sides with ψ as the interval, with R pixels as the radius, draw n line segments P1 to P n The score evaluation function of each line segment is: where Pixel(P i ) represents the total number of pixels that the line segment passes through. A path point direction of the optimal path is introduced as a second evaluation function, and a total score is obtained in combination with the first evaluation function, and an optimal path is selected for movement; this method can effectively complete local path planning and obstacle avoidance operation of the unmanned vehicle in the wild when part or all of the front is an impassable or obstacle region, specifically: If none of the n line segments in the current aerial view contacts the impassable region or the obstacle region, that is, the score evaluation function of each line segment is the same, then the mobile platform continues to move forward; when the line segment detects the impassable region or the obstacle region, the line segment is contracted inward along the center to the edge of the obstacle, at which time the direction of the longest line segment is determined, which is the direction in which the mobile platform needs to rotate and move; if there are several line segments with the same length, then the line segment is extended outward until it reaches the impassable region or the obstacle region or the boundary of the detection image, at which time the longest line segment is selected, that is, the direction indicated by the line segment with the maximum score evaluation function is selected as the direction in which the mobile platform needs to turn; if the same length still exists after extension, then another evaluation standard is introduced, that is, a path point direction score function is introduced as a second evaluation function for the optimal path, and a total evaluation function is obtained in combination with the first evaluation function, and the evaluation function of the path direction score is: where ε is a small value, ψ GPS represents the direction of the next path point, ψ i is the direction of the current mobile platform, and weight coefficients a and β are introduced, and the overall score evaluation function is: Score i = αScore Path,i + βScore Dir,i At this time, the optimal path is selected for movement according to the highest score, the mobile platform moves in this direction, and local path planning is completed.

2. The method of claim 1, wherein, The depth camera is installed at the central position in front of the unmanned vehicle, and an external computer is used to acquire a semantic segmentation result of a non-structured road in the wild by using a HRNet+OCR semantic segmentation network.

3. The method of claim 1, wherein, The semantic segmentation result is subjected to inverse perspective transformation to an aerial view and is subjected to binary processing, and a corresponding position of the unmanned vehicle in the aerial view is calculated, specifically: The semantic segmentation result of the non-structured road in the wild is subjected to inverse perspective transformation and is converted to an aerial view, and then binary processing is performed; in the image after binary processing, the four image points and the four object points are selected in line symmetry with respect to the camera shooting picture; for the aerial view result, the bottom midpoint is regarded as the current position of the depth camera on the mobile platform, so as to determine the corresponding position of the unmanned vehicle in the aerial view.

4. The method of claim 3, wherein, The center line of the aerial view is regarded as the direction pointed to by the front of the mobile platform.

5. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the steps of the method of any one of claims 1-4 when executing the program.

6. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program implements the steps of the method of any one of claims 1-4 when executed by the processor.

7. A computer program product comprising a computer program, characterized in that, The computer program implements the steps of the method of any one of claims 1-4 when executed by the processor.

Images