EAI-YOLO：Faster, and More Accurate for Real-Time Dynamic Object Detection

In sports event scenarios, precise detection of dynamic objects is crucial for real-time monitoring and event analysis. However, traditional dynamic object detection methods face challenges such as small target size, similar appearances, disordered motion, and dense occlusions, which often lead to problems like false detections, low accuracy, and poor robustness. To address these challenges, this paper proposes a real-time dynamic object detection algorithm EAI-YOLO based on improved YOLOv8s. First, the Efficient Channel Attention (ECA) is introduced to enhance the capability of capturing detailed features of small-scale moving targets. Second, the traditional Path Aggregation Feature Pyramid Network (PAFPN) is replaced by an Adaptive Feature Pyramid Network (AFPN) to improve the detection performance for densely occluded objects. Finally, a scale-adaptive Inner-CIoU loss is designed to enhance the detection generalization ability for blurry targets and those with abrupt scale changes. Experimental results show that on the self-made dataset, EAI-YOLO achieves an mAP@0.5 of 83.7% and an mAP@0.5–0.95 of 45.2%, with an inference time of only 0.034 seconds per frame, balancing real-time performance and detection accuracy. This study overcomes the issues of insufficient shallow feature extraction and performance degradation in occluded scenes of existing methods in complex motion scenarios, providing a high-precision technical solution for scenarios such as real-time sports event monitoring, athlete motion analysis, and intelligent referee systems. It has important application value for promoting the construction of "Smart Venues" and the intelligent development of competitive sports.