Research of Robust Video Object Tracking Algorithm Based on Jetson Nano Embedded Platform

Abstract

The majority of recent work has tended to improve the overall tracking capability of trackers. Despite the success of these methods, their inherent complexity has limited their scope of application. In contrast, it is more practical to improve the performance on embedded platforms without increasing the complexity of tracking algorithms. Limited by the computing power of the embedded system, this paper proposed an interesting tracker based on the Kernel correlation filter (KCF). To tackle the occlusion and disappearance problem during object tracking, we suggest a lightweight object detection network to relocate the target by using color features and then reinitialize the tracker. In addition, we avoid the problem of scale variation in object tracking by adaptively cross-linking the KCF algorithm with the embedded platform. The proposed method improves the performance of KCF tracker on embedded platform without increasing its complexity and achieves robust tracking of video targets. We have captured a lot of videos and done a lot of experiments to prove the effectiveness of our method.