一种组合核函数的自适应目标跟踪算法

Multi-feature description of adaptive kernels Object tracking

摘要:

分析了传统Mean Shift跟踪算法在外观模型对光照变化敏感以及外观模型更新上容易积累误差等缺点，结合了传统Mean Shift 跟踪算法计算速度快和易于组合的优点，设计了两种不同外观建模的Mean Shift跟踪算法。第一种Mean Shift跟踪算法采用传统的RGB颜色模型提取外观模型，第二种采用对光照变化不敏感的非色彩与梯度信息提取外观模型。结合这两种跟踪算法，通过这两种跟踪算法跟踪的目标进行加权得到的目标位置，以及根据协同更新的原理对这两种跟踪器的外观模板进行更新。这样不仅使得跟踪准确率得到了一定的提高，而且对外观变化的适应能力也大大的提高。

Abstract:

Analyzed the traditional Mean Shift tracking algorithm in appearance model is sensitive to illumination changes and the disadvantages such as easily accumulated error on appearance model updating, combines the traditional Mean Shift tracking algorithm calculation speed is fast and easy to combination, the advantages of the design of the two different appearance modeling Mean Shift tracking algorithm. The first kind of Mean Shift tracking algorithm using traditional RGB color model to extract the appearance model, the second is not sensitive to illumination change of color and gradient information extraction model appearance. Combining these two tracking algorithm, through the two track of target tracking algorithm weighted target location, and the principle of based on the update of the cooperation of the two kinds of the appearance of the tracker template updates. This not only makes the tracking accuracy has been improved, and the ability to adapt to change the appearance is greatly improved.