基于角点检测的实时目标跟踪方法

摘要

当目标区域与背景特征接近时,由于目标模型中包含了大量的背景信息,基于Mean Shift的经典跟踪算法易受到背景信息的干扰,造成目标定位偏差,甚至丢失目标.为此,提出了一种基于角点的实时目标跟踪方法.该方法利用SUSAN角点检测算子提取目标区域中的角点,由于目标区域内角点具有较强的目标表征能力,当使用这些角点构建目标模型时,能够增强目标与背景区域之间的辨别力,所以能够削弱背景信息对目标定位的干扰.实验结果表明,提出的方法能够实现对目标的准确跟踪定位,与经典的Mean Shift跟踪算法相比,该方法能够达到更好的跟踪效果,有效地提高了目标跟踪的准确性和实时性.%When the tracking target and the background have similar features, because the classic Mean Shift target model contains also background information, the tracking is easily affected by the interference of background, resulting in target location error and even loss of the target. To solve this problem, a real-time object tracking method based on corners is presented. The method uses SUSAN algorithm to extract corners in the tracking area. Because the corners can effectively represent the target, which increase the discrimination of the target and background, and target model based on these corners can reduce the impact of background information. The experimental results compared with classic Mean Shift tracking algorithm show, the proposed method can better track the target, thereby effectively improving the accuracy and the speed for real-time tracking.