摘要
视觉跟踪是当前计算机视觉的热点问题之一。TLD(Tracking Learning Detecting)算法是一种可以在线学习的新颖视觉跟踪算法。针对算法中跟踪器采用的LK光流法无法捕捉大幅度运动目标的问题,引入图像金字塔模型,提出一种采用金字塔光流法的TLD的改进算法,解决了长时间跟踪中出现运动尺度过大时产生孔径的问题。实验结果表明,算法在复杂场景和大运动条件下,可以长时间准确、快速地实现视觉跟踪,具有较强的适应性和有效性。
Visual tracking is a research hotspot in computer vision. TLD（Tracking-Learning-Detecting） algo- rithm is a novel visual tracking algorithm which can online learning. Aiming at the problem that LK flow method can not capture a large scale movement of TLD algorithm, an improved method is presented which used pyramid optical flow as the tracker, to get more comer information to overcome the aperture problem. The experimental results show the presented algorithm is efficient and robust to the large scale movement during long term tracking.
出处
《科学技术与工程》
北大核心
2013年第9期2382-2386,共5页
Science Technology and Engineering
基金
国家自然科学基金(61201378)
中央高校基础科研基金(N110804005)资助
