DESIGN AND IMPLEMENTATION OF BADMINTON LANDING POINT JUDGMENT DETECTION BASED ON YOLOv5

Based on the investigation and analysis of you only look once v5 (YOLOv5) target detection model, this paper proposes a detection algorithm optimized for badminton motion characteristics. YOLOv5 model was used for data training, and the Makesense tool was used to annotate the irregular rectangle in the data set. The convex hull algorithm is used to convert the COCO JSON format into YOLOv5 format, which optimizes the accuracy of annotation and improves the adaptability to high-speed motion, residual shadows and complex backgrounds of badminton. Through two rounds of incremental training and multiple optimizations, the badminton landing point detection algorithm is finally implemented, which significantly improves the recognition accuracy and robustness of the model. The system platform uses Python language, and the training and optimization of YOLOv5 model is realized based on Pytorch framework. The system includes data preprocessing, model training and evaluation, landing point prediction and other functional modules, and the running environment is Linux platform. The test results show that the system can accurately detect the badminton in complex scenes and predict its landing point in real time, demonstrating high real-time performance and accuracy for intelligent badminton analysis.