基于基础颜色特征的自适应尺度的多目标跟踪算法

在多目标跟踪领域，多个相似目标间相互遮挡时，易产生误跟踪、漏跟踪等问题。针对上述问题，通过引入语言学中的基础颜色及自适应尺度因子来解决。采用颜色命名过程及主成分分析法，提取目标基础颜色特征，准确区分相似目标；同时引入自适应尺度因子，自动改变目标尺度，减少因尺度变化而引入的干扰信息，增强目标外观模型的鲁棒性。基于以上两点，在Structure Preserving Object Tracking（SPOT）算法基础上，提出了Basic Color Adaptive Scale SPOT（CSSPOT）算法。在对比实验中，CSSPOT算法在跟踪准确率及计算时间这两方面较原算法均有所提升，充分说明了基础颜色特征及自适应尺度因子的正确性及有效性。

Multi-object tracking based on the basic color feature and adaptive scale

In multi-object tracking, when the occlusion between similar objects happen, the tracking error and drain tracking come up easily. To solve this problem, the basic color in the linguistic study had brought into multi-object tracking. Extracted the basic color feature of the target, through the process of color naming and PCA, can distinguish similar objects accurately. The adaptive scale factor had been considered in the meantime. This procedure can automatically changes the scale of the target to improve the robustness of the appearance model. Based on the Structure Preserving Object Tracking algorithm, proposed the multi-object tracking algorithm, Basic Color Adaptive Scale SPOT. The experiment shows that the CSSPOT has better performance both in the accuracy of tracking and the computing time than SPOT. Demonstrates the accuracy and effectiveness of basic color feature and adaptive scale factor.