Key points are not available for this paper at this time.
One of the most well-liked sports in the world now is badminton. Though their lack of intelligence and feedback mechanism restricts their application, traditional badminton serve machines can offer opportunities for practice. In order to identify, track, and anticipate the trajectory of badminton balls in flight, we therefore concentrate on the vision system of badminton robots. To identify badminton balls, we fuse the classical three-frame differencing with the machine learning technique AdaBoost. We also create a detection-based fast tracking approach for object centers (FTOC). In order to achieve even greater computational efficiency and more precise badminton ball coordinates, we enhance the Tiny YOLOv2 deep learning one-stage detection network and present the M-YOLOv2 and YOLOBR badminton ball detection networks.These enhanced networks perform exceptionally well in terms of memory, accuracy, and real-time badminton ball identification. YOLOBR testing yielded average precision, average recall, and average frame rate of 96.7%, 95.7%, and 29.2 frames/sec for four video streams of flying badminton in easy and challenging circumstances. According to the findings, YOLOBR outperforms Tiny YOLOv2 and M-YOLOv2 in terms of badminton identification accuracy, recall, and real-time performance. It also exhibits great scene robustness.
Haiyan Wang (Sat,) studied this question.