Persistent Target Tracking in Wide Area Using Multi-sourced Videos
Abstract：With the ever growing requirement for safety, videos captured by users, namely crowd-sourced videos along with CCTV videos are used as strong evidences for forensic applications. In this proposal, we aim at utilizing multi-sourced video data including CCTV videos and crowd-sourced videos for persistent target tracking in wide area. Current approaches of persistent tracking are based on multiple, fixed CCTV cameras which are not applicable if multi-sourced data are used. In this proposal, we present a comprehensive study of persistent target tracking using multi-sourced data. We fully utilize geospatial metadata of videos and propose a geospatial image filtering tool to select the most relevant video segments for target tracking. We also exploit the rich spatio-temporal context information in the video to enrich the representation of the target. The overall performance of persistent tracking is improved in terms of scalability, efficiency and tracking accuracy. This project will provide a solution to video surveillance in the new era of big data, which is both highly theoretical and has great potentials in real applications.
Keywords: computer vision; target tracking; mobile video; multiple camera; data management
Contact:
CAI Yinghao
E-mail: yinghao.cai@ia.ac.cn
Research Center of Aerospace Intelligent Information