稀疏编码目标跟踪方法研究

Research on sparse coding object tracking method

针对稀疏表示用于目标跟踪时存在重构误差表示不够精确、目标模板更新错误等问题,提出一种改进的稀疏编码模型。该模型无需重构误差满足特定的先验概率分布,且加入对编码系数的自适应约束,可以取得更优的编码向量,使得跟踪结果更为准确。在此基础上,将这种改进的编码模型与粒子滤波目标跟踪算法相结合,研究并实现一种新的基于鲁棒稀疏编码模型的目标跟踪方法。该方法对每个粒子的采样区域进行编码,用所得的稀疏编码向量作为当前粒子的观测量,并采用目标模板分级更新策略,使得目标模板更加准确。实验结果表明,方法可以较好地解决目标部分遮挡和光照变化等干扰下的目标跟踪问题。

To deal with the problems in sparse presentation’s application to object tracking, such as the not accurate enough representation of construction error, the wrong update of object templates and so on. This paper proposes a modified sparse coding model. This model doesn’t require the construction error following a particular prior probability density function, and adds an adaptive restraint of the coding coefficient, and this model can acquire better coding vectors for more accurate tracking results. Combined the modified model with the particle filter object tracking algorithm on this basis,a new object tracking method based on the robust sparse coding model is researched and realized. The proposed method encodes every particle’s sample area, uses the sparse coding vector as the observation for the particle, and updates the object templates with different grade to make them more accurate. The experiment results show that the proposed object tracking method can deal well with the corruption like occlusion and illumination variation in object tracking.