相关滤波器在视觉目标跟踪中得到了广泛应用, 针对复杂场景下目标跟踪容易出现跟踪漂移的问题, 以及现有多尺度跟踪算法计算量大的问题, 本文提出一种实时的多尺度目标跟踪方法。首先由时空上下文模型输出目标位置置信图完成目标定位, 再在尺度空间上训练相关滤波器完成目标尺度估计, 最后基于目标位置和尺度提出了一种新的时空上下文模型更新机制, 避免了模型更新错误。实验表明: 该方法在尺度变化、局部遮挡、目标姿态变化等情况下均能完成鲁棒跟踪, 跟踪正确率较原始时空上下文跟踪算法提高了 38.4%。

Correlation filter has been great applied in visual tracking. Aiming at the drift problem in complex situations, a multiple scale tracking method is proposed. Firstly, spatio-temporal context model is used to output the precise location of object by the confidence map. Secondly, the scale estimation is obtained by a trained correlation filter. Finally, based on the new location and scale, a new update mechanism of the spatio-temporal context model is proposed. Experimental results show that the proposed algorithm can complete the robust tracking under the condition of scale changes, partial occlusion, pose variations, etc. Tracking precision is improved by 38.4% compared with the original spatio-temporal context tracking method.