Luzhen Wang
Basketball robot, object detection, operating system, YOLOv5, instance normalisation
Basketball robots play a crucial role in autonomous collaboration and decision-making within games, but real-time object detection remains a challenge. This study analyses the framework of the robot operating system and introduces the YOLOv5s algorithm for enhanced laser detection. To improve learning and generalisation abilities, an optimal instance normalisation method is incorporated into the residual module. The proposed improved YOLOv5 model achieved an average accuracy of 0.932 after implementing ResUnit-b normalisation in the Backbone residual module, surpassing other algorithms by at least 5%. In addition, the improved YOLOv5 algorithm met the threshold specification requirements for both angle and distance after detecting all three spherical targets, with a minimum single-frame inference time of 23.6 ms (42 frames/s). Notably, its detection accuracy was 97%, the highest among similar models. In summary, the new object detection model significantly enhances the performance of basketball robots, contributing to advancements in the technology.
Important Links:
Go Back
