Robust Tracking via Color Names and Spatial Context Learning

Although visual tracking is a well-established problem in computer vision, it still remains a challenging task due to many disturbed factors such as appearance changes, illumination changes, partial or full occlusions, rotation, etc. Recently, spatio-temporal context (STC) learning algorithm has been proposed, which exploits the spatio-temporal relationships between the object of interest and its locally dense contexts. However, STC just simple models the spatial correlation between the center point of target and its surrounding regions with the simple low level features (image intensity and position), which ignores the importance of the appearance of the target. In our proposed algorithm, we add color names (CN) which represents the color appearance of the target into the intensity-based STC tracker, take the appearance of the object and the context into consideration simultaneously. Experimental results show that our tracker significantly outperforms several prior art approaches and the STC algorithm on various benchmark videos.