Pedestrian Tracking Combined with Deep Learning and Camera Network Topology in Non-overlapping Multi-camera Surveillance

摘要

This paper proposes a method of pedestrian tracking combining deep learning and camera network topology in non-overlapping multi-camera surveillance. The cross-camera pedestrian tracking task is divided into three parts: firstly, extracting structured information from single camera through deep learning; secondly, establishing the camera network topology; finally, extracting the candidate pedestrian in adjacent cameras by structured information and camera network topology, getting trajectory in next camera by using person re-identification method until there is no trajectory. Meanwhile, the experimental results show that our method can obtain pedestrian trajectory successfully and effectively.