A Multi-model Object Tracking Algorithm based on Texture and Color Features

Abstract

Mean Shift is widely concerned because of its advantages like fast convergence, real-time performance and simple procedure. However, the tracking performance of the traditional Mean Shift algorithm is obviously interfered when the background has the similar color as the target or the illumination changes. Besides, the tracking is easy to defeat in the case of target occlusion and loss of important frames. Therefore, this paper proposes two improvements based on the traditional Mean Shift tracking algorithm. First, the HLBP texture feature and color feature are employed to describe the target feature in order to improve the robustness of the tracking algorithm. Second, multiple models are taken into account to provide more abundant choices for the tracking process which can improve the tacking performance. The result of experiments show that our algorithm is more robust under the case of object occlusion and posture change, and gets better performance in accuracy when the background color and target color are similar or the illumination changes.