Robust Visual Tracking Using Illumination Invariant Features in Spatial Temporal Context Aware Model

Abstract

Visual tracking is one of the most active research topics due to its wide range of applications such as motion analysis, activity recognition, surveillance, and human computer interaction. The main challenge for robust visual tracking is to handle large appearance changes over time of the target object and the background due to occlusion, illumination changes, and pose variations. Numerous tracking algorithms have been proposed in the existing literature. In case of objects with appearance degraded severely, it becomes very difficult to extract enough useful tracking information from the target. The existing scene can therefore be a useful source of information to help localize the object. In other words, the statistical analysis of correlation between two frames of a video can effectively be used to determine the exact location of the target object. The existing algorithms do not fully exploit a strong spatio-temporal relationship that often exists between two consecutive frames of a video sequence. This thesis proposes a simple yet fast and robust algorithm, which exploits the dense spatio-temporal context for visual tracking. The approach formulates the spatio-temporal relationships between the object of interest and its local contexts in a Bayesian framework which models the statistical correlation of low-level features like image intensity and position from the target and its surrounding regions. Our experimental results on tracking benchmark video sequences show that the proposed method works favorably as compared to other state of art trackers. The proposed algorithm outperforms in terms of accuracy and robustness where illumination variation occurs dramatically.