Football education plays a significant role in university education, enriching campus life and promoting holistic student development. However, university football training in China faces challenges due to limitations in popularity, training resources, and environmental conditions. This study addresses issues of decreased detection accuracy caused by occlusion, uneven lighting, and small object size in traditional object detection systems by improving the YOLOv5 network structure. The original CSPDarkNet53 backbone was streamlined to a Mobile Net structure with depth wise separable convolutions, reducing model parameters and improving detection speed. Attention mechanisms were integrated into the YOLOv5 network to enhance target feature extraction and mitigate background interference, addressing occlusion and complex backgrounds. Additionally, the Unscented Kalman Filter was introduced into Deep SORT, replacing IoU with DIoU and employing a cascade matching method, significantly reducing ID switching in target tracking tasks. Experimental results on public datasets demonstrate that the proposed model exhibits superior detection performance, making it suitable for university football training scenarios. This study also highlights the potential of artificial intelligence, particularly object detection technology, to advance the efficiency and effectiveness of football training, contributing to the comprehensive development of student-athletes