Improved YOLOv7 Fusion Detection Line for Swimming Pool Drowning Detection

In order to ensure the safety of swimmers, an improved method for preventing drowning in video surveillance is proposed by integrating a region-based timing detection line into the YOLOv7 algorithm. The utilization of the region-based timing detection line enables preliminary assessment of swimmer behaviors, thereby reducing the rate of false drowning detections. To enhance the extraction of swimmer behavior features, the YOLOv7 algorithm incorporates the DCNv3 (Deformable Convolutional Network) module. Additionally, a Wise-IoU dynamic focal boundary box regression loss function based on attention is employed to decrease the rate of missed detections and enhance the overall detection accuracy. Experimental results demonstrate that the region-based timing detection line can accurately detect swimmers. Furthermore, the proposed algorithm exhibits a 3.6% improvement in recall rate and a 2.9% increase in average precision when compared to the YOLOv7 algorithm. Notably, it demonstrates superior performance in detecting behaviors of swimmers who are obstructed or located at a distance.