Visual Object Tracking Method of Spatio-temporal Context Learning with Scale Variation

Abstract

In the development of image processing and computer vision field, visual tracking is considered as an attractive research field in regard to its practical characteristic in security surveillance, computer–human based interaction, motion and activity recognition in health care or control systems, etc. In a typical visual tracking model, the most difficult task is to handle the changes in the target objects’ appearances and their surrounding backgrounds. As a matter of fact, if the changes are severe, information extracted to detect the object in interest will be limited. In this paper, we enhance the robustness of a tracking model to adapt to these changes and increase the tracking accuracy level by exploiting local context information. In particular, the study implements an efficient tracking model that utilizes the spatio-temporal context information. The context model relation of the tracking target with its surrounding background is generated by computing a devolution task due to its spatial correlation. Then, the analyzed relationship is exploited to update a spatio-temporal context in subsequent frames. The tracking process is computed using a confidence map by integrating the information within the spatio-temporal context. The model also implements the exhaustive scale estimation method to calculate the target’s scale characteristic changes while maintaining computational efficiency. Finally, the TB-100 dataset is applied to evaluate the performance of the model.