원문정보
초록
영어
Tracking-Learning-Detection (TLD), also known as Predator, has become one of the most popular state-of-the-art algorithms in the domain of visual tracking in recent years. It has demonstrated outstanding performance in the application of long-term tracking of a human face in unconstrained videos. In this paper, we address the problem of tracking a single ship in inland waterway CCTV videos given its location in the first frame and no other prior information. Firstly, we deeply analyze Predator in a perspective of control system and point out the search strategy in detection procedure is the most time-consuming part in Predator system. Secondly, in order to speed up the whole pipeline, we propose a novel motion model based on extended particle filter with orthogonal design. Due to the power of particle optimization and re-combination with orthogonal design, we can relate the motion of object of interest better and obtain the most likely candidates of object regions more effectively. Finally, both qualitative and quantitative evaluations on numerous challenging CCTV videos demonstrate that the proposed algorithm achieves favorable performance in terms of efficiency and accuracy.
목차
1. Introduction
2. Predator Closed-loop Control System
2.1. Predator Overview
2.2. Closed-loop Control System
3. Extended Particle Filter Motion Model with Orthogonal Design
4. Experiments
4.1. Qualitative Evaluations
4.2. Qualitative Evaluations
5. Conclusions
Acknowledgements
References