基于形状上下文和粒子滤波的多目标跟踪

Multiple target tracking using shape context features and particle filter

目标跟踪是计算机视觉领域里研究的热点和难点。提出一种基于形状上下文和粒子滤波的多目标跟踪算法,通过在跟踪过程中融入目标检测信息来处理目标进入与离开场景问题和目标重叠与分离问题。首先,采用自适应增强检测算法对视频区域中的目标进行检测;然后,利用形状上下文特征来建立被跟踪目标的外观模型;最后,利用粒子滤波方法进行粒子的选择和目标的跟踪。实验证明,提出的算法能够有效处理目标进入与离开场景的问题和目标重合与分离的问题,在单一背景和复杂背景下都能进行较为准确的跟踪,还能一定程度上处理部分遮挡问题。

Target tracking is still a hot and difficult research topic in computer vision. In this paper, we proposed a novel multiple target tracking method, which is based on shape context features and particle filter. Our method is proposed to deal with problems of the target objects entering （or leaving） the scene and overlap （or separation） by incorporating Adaboost detection. First, we adopt Adaboost detection algorithm for detecting multiple objects. Then, the target appearance model is builded by using the shape context features, Finally, we apply particle filter to choosing target particles and tracking objects in videos. The experiments indicate that the proposed method can effectively deal with the issue of targets entering（or leaving） the scene and targets overlap（or separation）, and exactly track the targets which is under both single and complex background. And the proposed method also can handle partial occlusion on some extent.