Intelligent golf ball picking vehicle path planning method based on visual detection

Abstract

The application discloses a kind of intelligent golf ball picking car path planning method based on visual detection, it is related to golf ball picking car technical field.The method includes: first, system initialization and global detection are carried out;Then GNSS positioning data and camera image are collected in parallel in driving;Identify whether golf ball exists in image;If exists, the pixel coordinates of ball are extracted;Discrete ball points are gathered into ball pile and the center point is calculated using DBSCAN density clustering algorithm;The relative position of ball pile is calculated by monocular ranging;Fusion GNSS data calculates the absolute coordinates of ball pile;Based on ball pile coordinates, the TSP problem is solved using greedy algorithm, to generate local optimal path and execute;Finally, it is judged whether global detection point is completed.The application realizes accurate ball picking in ball area by combining visual recognition and density clustering, avoids invalid driving, greatly improves work efficiency and energy utilization rate, and integrates mass discrete targets into a small number of ball piles, optimizes the operation path.

Description

Technical Field

[0001] This invention relates to the field of golf ball retrieval cart technology, and more specifically to a path planning method for an intelligent golf ball retrieval cart based on vision detection. Background Technology

[0002] Currently, golf ball retrieval at golf course training or practice ranges primarily relies on driverless ball-collecting vehicles. These vehicles typically employ high-precision RTK positioning technology, using pre-defined electronic fences for global path planning to achieve seamless, full-coverage navigation. However, this operational model has the following drawbacks: The lack of intelligent recognition of ball-collecting areas results in low operational efficiency. The vehicle can only travel along a preset path and cannot detect whether there are golf balls in the current working area. When the vehicle travels into an area without balls, it will still make ineffective travel, which not only reduces ball-collecting efficiency but also wastes limited battery life. The lack of a logical integration scheme for scattered targets means that, although some golf carts have introduced visual recognition technology, most can only detect targets and cannot convert the discrete pixels captured by vision into effective logical units in a geographic coordinate system. Due to the lack of spatial clustering analysis of golf ball density, the system cannot identify "ball clusters" and determine their geometric centers. This results in an excessive computational load on the backend algorithm when dealing with hundreds or thousands of scattered targets, making it difficult to plan the globally optimal operation path using an efficient greedy algorithm, leading to chaotic ball-collecting logic and frequent turning around.

[0003] Based on this, this application proposes a path planning method for an intelligent golf ball retrieval vehicle based on visual detection. Summary of the Invention

[0004] The purpose of this invention is to provide a path planning method for an intelligent golf ball retrieval vehicle based on visual detection, which combines visual recognition with density clustering to achieve accurate and efficient operation of the retrieval vehicle.

[0005] The technical problem solved by this invention is: (1) To solve the problem of inefficient "blind picking" operation caused by the lack of target recognition capability of existing golf carts, by introducing real-time image acquisition and visual recognition technology from cameras, the golf cart can intelligently sense whether there are golf balls in the working area, and eliminate the invalid driving in areas without balls. (2) To solve the problem of computational overload and logical confusion in path planning under a large number of scattered targets, the DBSCAN density clustering algorithm is used to integrate discrete golf balls into several "ball piles" with geometric centers, and the greedy algorithm is used to solve the traveling salesman problem to plan the globally optimal coherent operation path.

[0006] The objective of this invention can be achieved through the following technical solutions: A path planning method for an intelligent golf ball retriever based on vision detection includes the following steps: S1: System Initialization and Global Detection After the system starts up, it loads the preset global path detection points, and the ball-collecting car begins to cruise according to the global plan, entering the global path detection process; S2: Parallel Data Acquisition During operation, the system performs the following positioning data acquisition and visual data acquisition in parallel: Location data acquisition: Real-time acquisition of the ball-collecting vehicle's current GNSS location data to determine the vehicle's absolute geographic coordinates; Visual data acquisition involves obtaining real-time image data of the front view through an onboard camera. S3: Target Recognition and Judgment The system performs image recognition processing on the acquired real-time images from the camera to determine whether a golf ball exists in front of the camera's field of view. If the result is negative, proceed directly to step S9; If the judgment result is yes, then proceed to the target location processing flow; S4: Visual Coordinate Extraction After identifying the golf ball target in the image, extract the pixel coordinates of all golf balls in the image; S5: Density Clustering Analysis The DBSCAN density clustering algorithm was used to analyze the extracted golf ball coordinates. A density threshold was set to divide the scattered golf balls into several ball clusters, and the center coordinates of each ball cluster in the image were calculated. S6: Monocular ranging and relative position calculation Based on monocular ranging technology, the actual physical distance and orientation of each ball pile center relative to the camera are calculated according to the coordinates of the ball pile center in the image, thus obtaining the relative position information of the ball pile. S7: Multi-target GNSS coordinate calculation The current GNSS data of the ball-collecting vehicle obtained in step S2 is fused with the relative position information of the ball piles obtained in step S6 to calculate the absolute GNSS geographic coordinates of the center points of multiple ball piles within the field of view. S8: Local Path Planning and Execution Based on the calculated GNSS coordinates of multiple ball piles, a local path planning problem is constructed. A greedy algorithm is used to sort the target points of multiple ball piles and generate the optimal ball picking traversal path. The GNSS coordinates of the next target ball pile are selected in sequence, and the ball-collecting vehicle is controlled to move to the target point to carry out the operation. This process is repeated until the local ball-collecting task within the current field of view is completed. S9: Global Path Completion Determination The system determines whether the ball-collecting vehicle has passed all global path detection points; If not, return to step S1 or step S2, continue to the next global detection point and repeat the above detection and ball picking process; If so, the task is considered complete and the system terminates.

[0007] As a further aspect of the present invention: in step S2, the positioning data acquisition adopts a differential GNSS positioning module to obtain the vehicle position with centimeter-level accuracy.

[0008] As a further aspect of the present invention: in step S3, the image recognition uses a deep learning-based target detection model, such as YOLO or SSD, to detect golf balls in real time.

[0009] As a further aspect of the present invention: in step S5, the density threshold of the DBSCAN clustering algorithm includes the neighborhood radius Eps and the minimum number of points MinPts, which are preset or adaptively adjusted according to the actual distribution of golf balls.

[0010] As a further aspect of the present invention: in step S6, monocular ranging is based on the pinhole imaging principle, and calculates the actual distance corresponding to the pixel coordinates by combining the camera intrinsic parameters, installation height, and pitch angle.

[0011] As a further aspect of the present invention: In step S7, the coordinate calculation formula is as follows: the target point GNSS coordinates are the sum of the vehicle GNSS coordinates and the eastward and northward latitude and longitude increments after Earth curvature conversion, wherein the relative position is obtained by conversion of distance and azimuth.

[0012] As a further aspect of the present invention: in step S8, the execution process of the greedy algorithm is as follows: starting from the current vehicle position, each time the nearest unvisited ball pile is selected as the next target point, until all ball piles are visited.

[0013] As a further aspect of the present invention: when performing local path planning, if the number of ball clusters exceeds a preset threshold, the ball clusters are then subjected to secondary clustering to generate higher-level regional centers, thereby further optimizing the path.

[0014] The present invention also provides a vision-based intelligent golf ball retrieval vehicle, comprising: a power module, an on-board main control board, a GNSS positioning module, a camera, and a walking mechanism; the on-board main control board is connected to the GNSS positioning module, the camera, and the walking mechanism respectively, and is used to execute the path planning method described above.

[0015] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described path planning method.

[0016] The beneficial effects of this invention are: (1) Significantly improves the efficiency and energy utilization of ball retrieval operations. This invention abandons the traditional "blind cruising" operation mode of ball retrieval vehicles. By introducing real-time image acquisition and visual recognition technology from cameras, it achieves "active perception" of the work area. The system can accurately determine whether there are golf balls in front, thereby eliminating invalid driving paths in areas without balls. This not only significantly increases the number of balls retrieved per unit time, but also effectively reduces the energy consumption caused by motor idling and extends the single-charge range of the ball retrieval vehicle. (2) The discrete targets were logically integrated and the path planning algorithm was optimized. In view of the characteristics of the scattered distribution of golf balls, this invention innovatively introduced the DBSCAN density clustering algorithm, which can transform the massive number of discrete pixels recognized by vision into "ball pile" logical units with geometric centers. This effectively solves the data processing problem caused by the overly scattered targets in traditional vision solutions. By simplifying hundreds or thousands of targets into a small number of key nodes, the computational load of the back-end processor is greatly reduced, enabling the system to quickly solve the TSP problem using a greedy algorithm and plan a globally optimal operation route with fewer turns, shorter paths, and logical coherence. (3) This invention improves the real-time performance and robustness of the system. By acquiring GNSS data and visual data in parallel and combining monocular ranging with coordinate fusion, the target position can be updated in real time to adapt to the dynamically changing field environment. At the same time, the DBSCAN clustering algorithm can automatically filter out noise points to avoid interference from falsely detected targets. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart of a path planning method for an intelligent golf ball retrieval vehicle based on vision detection according to the present invention; Figure 2 This is a schematic diagram illustrating the detection and clustering effect in a vision-based intelligent golf ball retrieval vehicle path planning method of the present invention; Figure 3 This is a hardware system block diagram of a path planning method for an intelligent golf ball retrieval vehicle based on vision detection according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1 refer to Figure 1 - Figure 3 This embodiment provides a path planning method for an intelligent golf ball retriever based on vision detection. The retriever hardware includes a power module, an onboard main control board, a GNSS positioning module, a camera, and a walking mechanism. The onboard main control board is an embedded computer running a Linux operating system and equipped with image processing and path planning algorithms. The path planning method described in this embodiment includes the following steps: S1: System Initialization and Global Detection After the system starts up, it first loads the pre-stored electronic fence of the court and global path detection points. Global path detection points refer to several key locations covering the entire court area to ensure that the vehicle can cruise to all areas where the ball may land. The ball-collecting vehicle begins to travel along the global path and enters global detection mode; S2: Parallel Data Acquisition During operation, the system performs the following tasks in parallel at a high frequency: The vehicle's current latitude and longitude coordinates (accurate to the centimeter level) are obtained through a differential GNSS module and used as the vehicle's absolute position. Real-time images are captured using the front-facing camera, with an image resolution of 1920×1080 and a frame rate of 30fps. S3: Target Recognition and Judgment The acquired images are input into a pre-trained YOLOv5 object detection model to detect whether the images contain golf balls; If at least one golf ball is detected, proceed to the next step; otherwise, jump directly to step S9 to determine whether the global detection is complete. S4: Visual Coordinate Extraction For each frame of the image where a ball is detected, extract the pixel coordinates (u,v) of the center point of all detection boxes and save them as the set of points to be processed, P={(u1,v1),(u2,v2),…}. S5: Density Clustering Analysis Apply the DBSCAN clustering algorithm to the point set P, pre-set the neighborhood radius Eps=20 pixels and the minimum number of points MinPts=3, and cluster the points that are close to each other into one class, with each class representing a ball pile. For each class, calculate the average coordinates of all points as the center point of the ball pile in the image, and output the set of center points of the ball pile C={(uc1,vc1),(uc2,vc2),…}; S6: Monocular ranging and relative position calculation Based on the camera calibration parameters (focal length f, principal point coordinates (u0, v0)), installation height H (e.g., 1.5 meters), and pitch angle θ (e.g., tilted downwards by 15°), the actual distance d and azimuth angle α of each ball pile center relative to the camera optical center are calculated using the pinhole imaging model to obtain the relative position information of the ball pile, i.e.: d=H / sin(θ+arctan((v-v0) / f)); The azimuth angle α is calculated based on the deviation between the u-coordinate and the principal point; S7: Multi-target GNSS coordinate calculation Assuming the vehicle's current GNSS coordinates are (Lat, Lon) and the heading angle is yaw, the relative position of the sphere stack (distance d, azimuth angle α) is converted into an eastward offset ΔE and a northward offset ΔN centered on the vehicle: ΔE=d sin(α+yaw); ΔN=d cos(α+yaw); Then, the offset is converted into latitude and longitude increments using the Earth's radius of curvature to obtain the absolute GNSS coordinates (Lat_i, Lon_i) of the sphere's center. S8: Local Path Planning and Execution The GNSS coordinates of all spheres within the current field of view are used as nodes to be visited to construct the TSP model; A greedy algorithm is used to solve the problem: Starting from the vehicle's current position, the nearest unvisited pile of balls is selected as the next target point each time, until all piles of balls have been planned. Then, the vehicle is controlled to travel to each target point in sequence to pick up balls. Each time a pile of balls is reached, the system updates the visited list and checks in real time whether all piles of balls have been completed. After completing all piles of balls within the current field of view, proceed to step S9.

[0021] S9: Global Path Completion Determination The system checks whether all global detection points have been traversed; If there are still uncompleted detection points, continue traveling along the global path and repeat the above process; If all tasks are completed, the mission ends and the vehicle returns to the starting point or charging station.

[0022] Example 2 This embodiment is basically the same as embodiment 1, except that the density threshold of DBSCAN in step S5 can be dynamically adjusted according to the actual situation. For example, when too many balls are detected, Eps is automatically increased to merge balls that are farther away, reducing the number of clusters and thus speeding up subsequent path planning.

[0023] Example 3 This embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in Embodiment 1.

[0024] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A path planning method for an intelligent golf ball retrieval vehicle based on vision detection, characterized in that, Includes the following steps: Step S1: System Initialization and Global Detection After the system starts up, it loads the preset global path detection points, and the ball-collecting car begins to cruise according to the global plan, entering the global path detection process; Step S2: Parallel Data Acquisition During operation, the system performs location data acquisition and visual data acquisition in parallel: Location data acquisition: Real-time acquisition of the ball-collecting vehicle's current GNSS location data to determine the vehicle's absolute geographic coordinates; Visual data acquisition involves obtaining real-time image data of the front view through an onboard camera. Step S3: Target recognition and judgment The system performs image recognition processing on the acquired real-time images from the camera to determine whether a golf ball is present in front of the camera's field of view. If the result is negative, proceed directly to step S9; If the judgment result is yes, then proceed to the target location processing flow; Step S4: Visual coordinate extraction After identifying the golf ball target in the image, extract the pixel coordinates of all golf balls in the image; Step S5: Density Cluster Analysis The DBSCAN density clustering algorithm was used to analyze the extracted golf ball coordinates. A density threshold was set to divide the scattered golf balls into several ball clusters, and the center coordinates of each ball cluster in the image were calculated. Step S6: Monocular ranging and relative position calculation Based on monocular ranging technology, the actual physical distance and orientation of each ball pile center relative to the camera are calculated according to the coordinates of the ball pile center in the image, thus obtaining the relative position information of the ball pile. Step S7: Multi-target GNSS coordinate calculation The current GNSS data of the ball-collecting vehicle obtained in step S2 is fused with the relative position information of the ball piles obtained in step S6 to calculate the absolute GNSS geographic coordinates of the center points of multiple ball piles within the field of view. Step S8: Local Path Planning and Execution Based on the calculated GNSS coordinates of multiple ball piles, a local path planning problem is constructed. A greedy algorithm is used to sort the target points of multiple ball piles, generate the optimal ball picking traversal path, and select the GNSS coordinates of the next target ball pile in sequence. The ball picking vehicle is then controlled to move to the target point to perform the operation. This process is repeated until the local ball picking task within the current field of view is completed. Step S9: Global path completion determination The system determines whether the ball-collecting vehicle has passed all global path detection points; If not, return to step S1 or step S2, continue to the next global detection point and repeat the above detection and ball picking process; If so, the task is considered complete and the system terminates.

2. The path planning method for an intelligent golf ball retriever based on vision detection according to claim 1, characterized in that, In step S2, the positioning data acquisition uses a differential GNSS positioning module to obtain the vehicle position with centimeter-level accuracy.

3. The path planning method for an intelligent golf ball retriever based on vision detection according to claim 1, characterized in that, In step S3, the image recognition uses a deep learning-based target detection model to detect golf balls in real time.

4. The path planning method for an intelligent golf ball retriever based on vision detection according to claim 1, characterized in that, In step S5, the density threshold of the DBSCAN clustering algorithm includes the neighborhood radius Eps and the minimum number of points MinPts, which are preset or adaptively adjusted according to the actual distribution of golf balls.

5. The path planning method for an intelligent golf ball retriever based on vision detection according to claim 1, characterized in that, In step S6, monocular ranging is based on the pinhole imaging principle, and calculates the actual distance corresponding to the pixel coordinates by combining the camera's intrinsic parameters, installation height, and pitch angle.

6. The path planning method for an intelligent golf ball retriever based on vision detection according to claim 1, characterized in that, In step S7, the coordinate calculation formula is as follows: the target point GNSS coordinates are the sum of the vehicle GNSS coordinates and the eastward and northward latitude and longitude increments after Earth curvature conversion, where the relative position is obtained by conversion of distance and azimuth.

7. The path planning method for an intelligent golf ball retriever based on vision detection according to claim 1, characterized in that, In step S8, the greedy algorithm is executed as follows: starting from the current vehicle position, the nearest unvisited ball pile is selected as the next target point each time, until all ball piles are visited.

8. The path planning method for an intelligent golf ball retriever based on vision detection according to claim 1, characterized in that, When performing local path planning, if the number of clusters exceeds a preset threshold, the clusters are clustered again to generate higher-level regional centers, thereby further optimizing the path.

9. A vision-based intelligent golf ball retrieval vehicle, characterized in that, include: The system includes a power module, a vehicle-mounted main control board, a GNSS positioning module, a camera, and a walking mechanism; the vehicle-mounted main control board is connected to the GNSS positioning module, the camera, and the walking mechanism respectively, and is used to execute the path planning method according to any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the path planning method according to any one of claims 1-8.

Images