YOLOv7-DCN行人检测研究

The YOLOv7-DCN Pedestrian Detection Study

由于受到监控视频中行人的遮挡、尺度变化、光照等影响,行人检测容易出现漏检和错检等问题。针对以上这些问题,提出了一种基于YOLOv7改进的YOLOv7-DCN行人检测算法,用可变形卷积块DCNV3替换CBS卷积块中的常规卷积块,将YOLOv7的损失函数替换为MPDIo U损失函数,在YOLOv7中加入Sim AM注意力机制。改进后的YOLOv7-DCN检测算法检测精准度m AP提升了3.6%,对行人有较好的检测效果。

Due to the occlusion,scale change,illumination and other issues of pedestrians in the surveillance video,pedestrian detection is prone to missed detection and false detection.In response to these problems,an improved YOLOv7-DCN pedestrian detection algorithm based on YOLOv7 is proposed.The conventional convolution block in CBS convolution block is replaced by deformable convolution block DCNV3,the loss function of YOLOv7 is replaced by MPDIoU loss function,and the SimAM attention mechanism is added to YOLOv7.The detection accuracy mAP of the improved YOLOv7-DCN detection algorithm is increased by 3.6%,which has a good detection effect on pedestrians.