基于多特征融合的压缩跟踪算法

压缩跟踪将压缩感知理论引入到目标跟踪领域，较好地实现了跟踪的实时性，但是在复杂环境或遮挡情况下，仅利用分类分数最大值的矩形样本确定目标位置容易产生跟踪漂移，而且该算法没有考虑目标尺度因素。针对这些问题，提出了融合局部中心区域梯度方向直方图和多尺度矩形的多特征压缩跟踪算法，并提出利用多样本矩形平均的方法确定最终的目标位置。实验结果表明：该算法在目标剧烈运动、遮挡或者相似物体干扰的场景下能够有效抑制跟踪漂移，提高了跟踪的准确率和鲁棒性。

Compressive tracking based on multi-feature fusion

Compressive sensing theory is applied into object tracking area for a compressive tracking algorithm. It can better realize real-time tracking, but it?can easily cause?drift by only using rectangle samples of maximum classification?score value in complicated environment or occlusion. Moreover, the algorithm does not consider the factor of target scale. It presents a multi-feature compressive tracking algorithm based on gradients oriented histogram of local center area and multi-scale rectangle for these issues. It also proposes final object position determination by the average of multiple rectangle samples. Experimental?results show that the algorithm can effectively suppress the tracking drift and improve the accuracy and robustness of a tracking algorithm in such scenarios as violent object moving, occlusion and similar object interference.