基于相关滤波器的长时舰船目标跟踪方法

Ship object tracking based on long-term correlation filter

在海上复杂环境下对目标进行跟踪时,会出现目标遮挡、尺度变化等问题。针对这些问题,引入长时间相关滤波算法,将跟踪任务分解为对目标平移和尺度的估计,其中目标平移通过时空上下文进行估计,而目标尺度通过搜寻外观金字塔进行估计。另外,使用在线随机蕨分类器对跟踪失败的目标再检测,从而实现长时间跟踪。将该算法首次应用于海上舰船跟踪中,通过与其他几种长时间跟踪算法进行仿真实验对比,结果表明:该算法跟踪精度和成功率均有较大提高。在目标发生遮挡和尺度变化等情况下,该方法均具有较高的精确度。

When the ship targets are tracked in the complex ocean background, the problems such as occlusion and the change of the target scale are presented. In order to overcome these problems, the long-term correlation filtering algorithm is introduced. In this setting, we decom- pose the task of tracking into translation and scale estimation of objects. The translation is estimated by modeling the temporal context correla- tion and the scale is estimated by searching the target appearance pyramid exhaustively. In addition, we train an online random fern classifier to re-detect objects in case of tracking failure, and to achieve long time tracking. The long-term correlation filtering algorithm is first applied to ship object tracking, and comparative experiments show that the proposed algorithm gets the better performance on distance precision and suc- cess rate. Also, it is robust to scale changing and heavy occlusion.