An Optimized Tennis Ball Picking Method Combining YOLOv5 and K-Means++
Peixiang Li, Haiyan Zhou, Xingyu Yang · 2024
With the rapid development of artificial intelligence and robotics, intelligent picking robots are being increasingly applied across various fields. However, how to efficiently identify and pick up the target object in a complex dynamic environment is still a challenging problem. Aiming at this problem, an optimized visual recognition and path planning method is developed by fusing YOLOv5 target detection algorithm and K-Means++ clustering algorithm to improve the performance of tennis picking robot. YOLOv5 is employed within the ROS framework to detect and locate tennis balls, while K-Means++ is used to cluster and optimize their positions. Field experiments demonstrate that the system achieves [email protected], [email protected]:0.95, accuracy, and recall rates of 98.26%, 86.14%, 99.14%, and 96.59%, respectively, in tennis ball recognition. The improved K-Means++ clustering algorithm shows strong performance in optimizing path planning. Compared to the unimproved K-Means algorithm, the average picking time is reduced by 12.14%. When compared to the Ant Colony Optimization, the reduction is 4.281%, and compared to the Genetic Algorithm, the picking time is decreased by 8.076%. Additionally, when compared to manual picking, the time is reduced by 19.589%.