Towards the design of smart video-surveillance system

Security and monitoring systems are increasingly demanding in terms of quality, reliability and flexibility especially those dedicated to video surveillance. The aim of this study is to identify some limiting factors in the existing video-surveillance systems and to propose a set of best practices for developing a smart platform for a security monitoring system incorporating advanced techniques for video processing and analysis. In this work, we focus on the effect of the video quality on the biometric part of the video-surveillance systems for public security. In such systems, face detection and recognition from video sequences acquired from surveillance cameras, are challenging tasks, due to presence of strong illumination variations, noise, and changes in facial expressions. In this paper, we mainly focus on the illumination issue occurred in video surveillance. The low light video data is processed using a perceptual based approach, namely multi-scale Retinex method, to improve the video quality, followed by face detection. The experimental results demonstrate significant performance improvement in face detection and recognition, by improving the illumination of video sequences over the unprocessed video data.