融合颜色直方图响应的时空上下文跟踪算法

时空上下文(STC)跟踪算法在特征表达、尺度自适应策略等方面存在缺陷，当出现目标突然形变、局部遮挡或尺度变化等情况时，跟踪器的性能会严重退化。通过对STC算法进行改进，提出了一种融合颜色直方图响应的时空上下文跟踪算法。基于颜色统计的模型对运动模糊和目标形变等影响因素不敏感，和时空上下文模型具有良好的互补性质，在响应层融合后能够提升算法的鲁棒性。此外，采用基于多尺度金字塔模型的尺度搜索策略替换STC算法中原有的尺度估计策略，进行更精准的自适应尺度估计。在大规模公开数据集上的测试结果表明，本文算法在不同影响因素的复杂环境下展现了更为良好的跟踪性能和适应性，并且平均跟踪速度达到134.2帧/秒。

Spatio-temporal context tracking algorithm for merging color histogram response

The Spatio-Temporal Context(STC) tracking algorithm has defects in feature representation and scale adaptive strategy. Undesired conditions, i.e. abrupt deformations, partial occlusions or scale variations of the object appearance, would severely degrade the performance of the tracker. In this paper, based on the improvement of the STC algorithm, an algorithm is proposed for merging template response according to the STC model and color histogram response to locate the target object. The color statistics-based model has a good complementary nature to the STC model. The STC tracker combining the color histogram response can be inherently robust to both motion blur and deformations. Moreover, another scale search strategy which is based on a multi-scale pyramid model is adopted to replace the scale module in STC tracker, and makes scale estimation more accurately and adaptively. Extensive experimental results on large-scale benchmark sequences show that the proposed algorithm exhibits better tracking performance and adaptability under the complex environment of different influencing factors while running at 134.2 frames/s on average.