基于视觉显著性和特征点匹配增强的运动目标跟踪框架

为了解决主流目标跟踪算法在目标遮挡、快速位移和光照变化等不利条件下易发生跟踪不稳定的问题,提出一种视觉显著性和特征点匹配增强的目标跟踪框架,当瞬时跟踪结果经判别达到SURF特征点匹配阈值时,触发跟踪增强机制,提取图像的时域、频域和空域显著性得到综合的显著图,然后执行邻近匹配策略搜索显著图各显著区域以找到目标最可能的存在区域.本文框架是在传统跟踪方法的基础上加入了独立的增强机制,因此并不局限于某一种特定跟踪方法,具有良好的可扩充性和通用性.实验结果表明,该框架能有效克服多种不利因素的带来的干扰,可实现复杂背景下目标的准确跟踪.

A target tracking framework enhanced byvisual saliency and feature points

A target tracking framework enhanced by visual saliency and feature points was proposed to cope with the impact derived from occlusion, fast-moving and illumination changing and so on. As the threshold of SURF feature points matching was reached, the enhanced tracking mechanism was triggered, and the saliency extracted from time domain, frequency domain and space domain was synthesized to obtain the final saliency map. A vicinity-matching strategy was proposed to find the most likely position of the target. As the proposed framework was formed by integrating the independent enhanced mechanism into traditional tracking methods, it was not limited to a certain tracking method. Comparative experiments show that the framework has the ability of overcoming the impact derived from various unfavorable factors, and the robust target tracking is performed in cluttered environment.