Abstract: As traditional fast tracking via spatio-temporal context learning fails to track target stably when target is in the complex background and occlusion condition, a robust fast tracking via spatio-temporal context learning is proposed. The Kalman filter is used to estimate and predict the target’s position in the next frame of current frame. The estimated position is used as the starting point of the iteration of the fast tracking via spatio-temporal context learning in the next frame. Results of tests on variant video sequences show that the proposed algorithm has advantages over fast tracking via spatio-temporal context learning and multiple instance learning tracking when target is in the complex background and occlusion condition. Obtained results satisfy the requirements of real-time tracking.

Key words: spatio-temporal context, drift, Kalman filter, multiple instance learning, real-time

摘要： 为解决时空上下文快速跟踪算法在目标处于复杂背景及被遮挡情况下容易产生漂移的问题，提出了一种鲁棒的时空上下文快速跟踪算法，通过引入Kalman滤波器，对当前帧中的目标在下一帧中的位置进行估计和预测，并将其作为下一帧时空上下文快速跟踪算法的迭代起点。对不同视频序列的跟踪结果表明，与时空上下文快速跟踪算法和多示例学习跟踪算法相比，提出的算法在目标被遮挡及复杂背景情况下能够更准确地跟踪到目标，并且满足实时性要求。

关键词: 时空上下文, 漂移, Kalman滤波, 多示例学习, 实时性