Research on dense crowd detection algorithm based on improved YOLOv8

Dense crowd analysis is significant in public security, intelligent monitoring, and other fields. However, when dealing with dense and overlapping targets, the traditional target detection algorithm will face the problems of overlapping target frames, reduced detection accuracy, and insufficient feature extraction. Therefore, this paper proposes an improved YOLOv8 model by introducing the MPDIoU loss function and SEAttention attention mechanism to improve the detection performance of the model in dense crowd scenes. MPDIoU loss function enhances the separation ability of the model for dense and overlapping targets by optimizing the average distance between target frames to improve detection accuracy. At the same time, the attention mechanism is embedded in the backbone network to enhance the correlation and importance weight between feature channels and effectively improve the expression ability of feature extraction. The experimental results show that the improved YOLOv8 model performs better than the original on multiple dense population datasets, especially when dealing with dense and overlapping targets. The method proposed in this paper provides an efficient and reliable solution for dense population analysis and has a broad application prospect.