Basic Soccer Movement Detection Based on YOLO v8

Abstract

The paper defines object detection and discusses the advantages and disadvantages of one-stage and two-stage detectors. A literature review demonstrates the excellent performance of You Only Look Once Version 8 (YOLO v8) in object detection and its application in soccer. Based on this, the paper uses YOLO v8 as a pre-trained model and leverages its object detection capabilities to detect basic soccer moves in images. With a manually collected and labeled training set, YOLO v8 is pre-trained on four moves: penalty kicks, shoots, dribbling and headers. Experiment results were obtained with a validation set. The results show that at a confidence level of 0.6, the average precision, the recall and the F1 score across all classes are approximately 0.78, 0.65 and 0.7 respectively. At the intersection over union ratio (IoU) threshold of 0.5, the mean average precision (mAP) across all classes is 78.5%, demonstrating good performance of the model. Future research may include human body modeling for soccer moves and human pose estimation to improve the detection accuracy of the model.