基于YOLOv8和DeepSort的多区域行人追踪算法研究

Research on Multi-area Pedestrian Rracking Agorithm Based on YOLOv8 and DeepSort

针对多摄像头重叠场景中行人追踪容易发生身份丢失、切换的问题,本文提出了一种基于YOLOv8和DeepSort的多摄像头跟踪算法.在检测阶段,利用无参注意力机制增强网络对行人特征的提取能力,提高了检测器的性能.在追踪阶段,通过提取两个摄像头的视角关键点,并计算出两个视角的单应性矩阵,实现了不同视角图像的拼接.通过利用目标间的单应性关系,在DeepSort算法中完成目标匹配.并在MOT15数据集中,对所改进的算法进行了测试.实验结果表明,本文提出的基于YOLOv8和DeepSort的改进算法的平均跟踪精确度为63.5%,比原始算法提升了3.4%.改进算法在行人身份切换次数方面减少了52次,比原始算法减少了6.5%.

In order to solve the problem that identity loss and switching are easy to occur in pedestrian tracking in multi-camera overlapping scenes,this paper proposes a multi-camera tracking algorithm based on YOLOv8 and DeepSort.In the detection stage,the parameter-free attention mechanism was used to enhance the network′s ability to extract pedestrian features,and the performance of the detector was improved.In the tracking stage,the key points of the viewing angles of the two cameras were extracted,and the mono-response matrix of the two viewing angles was calculated,so that the stitching of images from different viewing angles was realized.By using the mono-response relationship between targets,the target matching is completed in the DeepSort algorithm.And in the MOT15 dataset,the improved algorithm was tested.Experimental results show that the average tracking accuracy of the improved algorithm based on YOLOv8 and DeepSort is 63.5%,which is 3.4%higher than that of the original algorithm.The improved algorithm reduces the number of pedestrian identity switches by 52 times,which is 6.5%less than the original algorithm.