Eagle-eye detection method for dynamic and static cross-line in football matches integrating Unet3 + and YOLOv5

Abstract The boundary line detection in football matches has developed from manual judgment to the integration of video assistant referees and electronic eagle-eye technology. Under static conditions, it relies on ultra high definition cameras to freeze and draw lines, while under dynamic conditions, it utilizes ball sensors and magnetic coils for detection. However, there are still problems such as limited dynamic detection range, inconsistent multi-source data fusion criteria, and process delays caused by human intervention. The research aims to build an efficient and accurate dynamic and static cross-line eagle-eye detection system for football matches to achieve fast and accurate penalty decisions in complex stadium environments. The eagle-eye detection system for football matches is designed. A full-size layer-hopping U-shaped segmentation network is employed to achieve accurate edge segmentation through full-size feature fusion and deep supervision. The YOLOv5 lightweight version of the baseline backbone network is replaced with Residual Network with 50 layers. The Squeeze-and-Excitation is integrated to calibrate feature weights. An efficient alignment detection head is used to optimize target positioning and combined with the background difference method to complete line crossing judgment. The results showed that the edge segmentation intersection over union reached 0.968, the average target detection precision was 0.965, the dynamic scene accuracy was 93.6%, and the detection speed was 52.8 frames per second. This detection method effectively breaks through the limitations of existing technology, provides reliable technical support for line-breaking penalties in football matches, and helps improve the efficiency and fairness of game penalty decisions.