A new method of video-surveillance data analytics for the security in camera networks

International audienceKeeping smart cities safe against acts of violence and security breaches is something critical. In a smart video-surveillance framework, human re-identification in non-overlapping camera networks represents a major challenge. To deal with the uncontrolled variation of the surveillance area such as viewpoint and illumination changes, and occlusion, it is essential to seek the most robust object model invariant during changes. By exploiting the performance of the spacetime covariance model, we propose a new approach based-on the analysis of all the video data extracted from camera-networks. This approach not only deals with one video-frame as the majority of methods, but also considers all the extracted groups of pictures to implicitly encode the described person in motion by the integration of time parameter and the clustering step. The experiments conducted on PRID-2011 database showed the importance of data analytics in recognition rates