基于活动状态预测与分类的多目标跟踪

固定单摄像机多目标跟踪的难点在于对多目标互相遮挡情况的处理。针对此问题，本文提出一种基于活动状态预测与分类的多目标跟踪算法，并通过对数字视频中多人的跟踪对算法进行测试。通过对实际情况的分析总结将目标活动状态分为六类，利用卡尔曼滤波对遮挡的预测信息，结合区域匹配信息对目标的活动状态进行归类。最后，通过采用
用不同的目标定位及模板更新策略处理不同活动状态的目标，达到跟踪的目的。实验证明，对目标的分类处理使算法对多目标跟踪具有较好的适应性和准确性。

Multiple Object Tracking Using Object Activity Prediction and Classification

Object occlusion is a difficulty in multiple object tracking in video with a fixed camera. A multiple object tracking algorithm is introduced based on the object activity prediction and classification for processing this difficulty. Object activities are classified into six classes. Then object activities are classified through moving region matching. Object occlusion is predicted using the Kalman filtering, and the objects in different status are processed by different methods of locating and template updating. Experimental results show the good adaptability of the algorithm in multiple object tracking.