Pedestrian multi-target tracking method based on feature association and feature enhancement

The invention provides a pedestrian multi-target tracking method based on feature association and feature enhancement, and relates to the technical field of multi-target tracking. According to the method, DLA34 is adopted as a target detection backbone network, a feature association module based on self-attention and mutual attention is added, a target detection result is obtained through combination of a Desection branch and a ReID branch enhanced in space and channel, a detection target is matched with a track through a DeepSort algorithm, and the method aims at effectively detecting and distinguishing pedestrians appearing in an image and improving the detection accuracy. The detection precision is improved, and the ID switching number is reduced. 本发明提供了一种基于特征关联与特征增强的行人多目标跟踪方法，涉及多目标跟踪技术领域。该方法采用DLA34作为目标检测主干网络、加入了基于自注意力和互注意力的特征关联模块，通过Detection分支和空间、通道增强的ReID分支相结合，得到目标检测结果，通过DeepSort算法将检测目标与轨迹相匹配，旨在能够有效检测区分图像中出现的行人，提升检测精度，降低ID切换数。