Intelligent recognition of fouls by football players based on human keypoint detection

To improve the accuracy of recognising fouls committed by football players, an intelligent recognition method based on human keypoint detection is proposed. First, machine vision technology is used to collect motion images of football players, and Gabor filtering is applied to process the football motion data. Next, the deep layer aggregation architecture (DLA) is employed to detect human keypoints in football player action images. By predicting keypoint heatmaps, local displacements, and target sizes, accurate human keypoint localisation is achieved. Finally, the Harris 3D operator is used to construct a potential energy model for fouls, combined with K-means clustering, BOW feature mapping, and the AdaBoost algorithm, enabling intelligent recognition of fouls by football players. Experimental results show that the average detection accuracy of human keypoints using the proposed method reaches 94.212%, and the recognition rate for foul actions remains above 90%.