Improved YOLO-based fusion attention mechanism dense crowd detection algorithm

The invention provides an intensive pedestrian detection algorithm based on an improved YOLO (You Only Look Once) fusion attention mechanism on the basis of deep learning. On the basis of a YOLOv5 pedestrian detection network, a CA (Coordinate Attention) attention mechanism module is introduced, CA is fused into a YOLOv5-C3 structure, the algorithm detection performance is improved while few parameters are introduced in the calculation process, and the omission factor of pedestrian detection is effectively reduced and the detection precision is greatly improved by adopting a method for optimizing a redundant candidate prediction frame by EIoU-NMS. 本发明以深度学习为基础提出一种基于改进YOLO（You Only Look Once）的融合注意力机制密集行人检测算法。以YOLOv5行人检测网络为基础，引入了CA（Coordinate Attention）注意力机制模块，将CA融合入YOLOv5-C3结构中，在计算过程中引入较少参数的同时提升了算法检测性能，并且采取了EIoU-NMS来优化冗余候选预测框的方法，有效降低了行人检测的漏检率，大大提高了检测精度。