关键词: 实时目标追踪, 统一的实时目标检测, 尺度不变特征转换

Abstract: This paper proposes a robust real-time tracking algorithm for weak targets based on SIFT and YOLO. It can continuously and steadily track the target under complex conditions such as dramatic scene changes and object occlusion.The implementation of the algorithm is based on a reasonable algorithm architecture design, uses YOLO to select candidate targets, and uses SIFT to select the tracked targets from the candidate targets.The tracking algorithm proposed in this paper not only meets the real-time performance, but also obtains better experimental results than KCF and CamShift in the normal test set, the target existence occlusion test set, and the camera rotation test set.The results show that the tracking algorithm proposed in this paper has outstanding performance in solving problems such as target occlusion and dramatic scene changes.

Key words: real-time target tracking, unified real-time target detection； scale invariant feature transform