Person Detection, Re-identification and Tracking Using Spatio-Color-based Model for Non-Overlapping Multi-Camera Surveillance Systems
Farah Jahan · The Smart Computing Review · 2012
The main goal of smart video surveillance is to develop an intelligent system in order to provide security to people and sensitive areas. Both person detection and tracking are challenging and crucial problems in intelligent video surveillance systems because a body changes its shape while moving. Multiple camera-based visual surveillance systems can be extremely helpful in expanding a surveillance area and avoiding occlusion. To monitor a wide area of interest, re-identification of persons across multiple disjointed fields of view is crucial. In this paper we propose a novel method for detecting, re-identifying and tracking persons as they move between different disjointed camera views. We use color thresholding to detect the presence of moving objects and validate them for personhood by analyzing shape information. We generate a discriminative signature for each person based on their spatio-color appearance information in order to re-identify them. The spatio-temporal information as well as labels obtained by our re-identification method helps the system to keep track of persons. Our experiment demonstrates that the proposed approach works in real-time and demonstrates excellent detection, re-identification and tracking accuracy.