| 摘 要: 研究了传统跟踪—学习—检测(Tracking-Learning-Detecting)目标跟踪算法的结构和特点，提出改进思 路；虽然TLD算法采用P-N学习机制，在应对长时间跟踪方面有很好的鲁棒性，但是当目标发生严重遮挡、形变，或者 场景发生较大的光照、旋转变化时，也会导致跟踪的失败。基于对以上问题的研究，提出TLD改进跟踪算法。改进算法 在跟踪模块运用SIFT特征匹配算法来代替原算法中LK光流法，减少了计算的复杂度，提高了算法的环境适应能力。 |
| 关键词: 目标跟踪；TLD算法；LK光流法；P-N学习；SIFT |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 国家自然科学基金重大仪器专项，项目编号：11527801；国家自然科学基金青年项目，项目编号：41706201. |
|
| Video Target Tracking Algorithm Based on Improved TLD Framework |
|
SHI Shufan,SUN Guangmin
|
( Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China)
|
| Abstract: In this paper,the structure and characteristics of the traditional tracking-learning-detecting (TLD) target tracking algorithm are studied and the improved idea is proposed.Although the TLD algorithm adopts the PN learning mechanism,it has good robustness in dealing with long-term tracking.However,when the target is severely occluded,deformed,or when there is a large illumination or rotation change for the scene,it may also lead to failure of tracking.Based on the research on the above-mentioned problems,an improved TLD tracking algorithm is proposed.The improved algorithm uses the SIFT feature matching algorithm in the tracking module to replace the LK optical flow method in the original algorithm,which reduces the computational complexity and improves the environment adaptability of the algorithm. |
| Keywords: target tracking;TLD algorithm;LK optical flow method;P-N learning;SIFT |
