Towards Real-Time Visual Tracking in Mobile Robots

This paper presents a robust and real-time target detection and tracking method under the framework of Meanshift for autonomous robot applications. Meanshift, because of its simplicity and efficiency, has been widely used in tracking tasks, especially in embedded environments. But, the traditional MS algorithm can be effective only when the target regions located in the previous frame are very similar to the candidate regions of the current frame. Therefore, the tracking performance will deteriorate when the target moves fast, or the target is blocked, because at this time, the overlapping area in the image will be greatly reduced. In this paper, we integrate the background models and the graded color features of moving targets to solve the above challenges. In addition, our method is tested on 10 video streams with different scenarios on the ARMv7 platform. Experimental results show the effectiveness of this method in the above scenarios.