EXTRACTING 3D SEMANTIC INFORMATION FROM VIDEO SURVEILLANCE SYSTEM USING DEEP LEARNING

International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences

Zhang, J. S.^1¹  Cao, J.^1² 

[1] Nanjing University of Finance & Economics, College of Information Engineering, Collaborative Innovation Center for Modern Grain Circulation and Safety, Jiangsu Key Laboratory of Modern Logistics, Nanjing, 210023, China^1;Nanjing University of Science & Technology, Nanjing, 210094, China^2

关键词: 3-D space;    Camera calibration;    Target recognition;    Target tracking;   

DOI  :  10.5194/isprs-archives-XLII-3-2257-2018

学科分类：地球科学(综合)

来源: Copernicus Publications

【 摘 要 】

At present, intelligent video analysis technology has been widely used in various fields. Object tracking is one of the important part of intelligent video surveillance, but the traditional target tracking technology based on the pixel coordinate system in images still exists some unavoidable problems. Target tracking based on pixel can’t reflect the real position information of targets, and it is difficult to track objects across scenes. Based on the analysis of Zhengyou Zhang's camera calibration method, this paper presents a method of target tracking based on the target's space coordinate system after converting the 2-D coordinate of the target into 3-D coordinate. It can be seen from the experimental results: Our method can restore the real position change information of targets well, and can also accurately get the trajectory of the target in space.

【 授权许可】

CC BY