一种基于超像素的局部判别式跟踪算法

Local discriminative tracking algorithm based on superpixel

针对目标在复杂环境下容易受到外界干扰而发生漂移的问题,提出了一种基于超像素的局部判别式跟踪方法.首先,对视频序列前10帧的目标区域进行分割,得到超像素,并利用kmeans方法对其进行聚类以构造初始字典;其次,通过训练样本集来训练线性分类器;然后,为了减少目标发生漂移的可能性,将初始训练的分类器与更新后的分类器线性加权之和定义为似然函数;最后,在粒子滤波的框架下,将似然函数值最大的粒子作为跟踪的结果,每运行U帧更新一次字典和分类器参数,以捕获目标表观的变化.仿真结果表明,所提算法在目标发生遮挡、光照变化的复杂环境下仍然能够跟踪目标.

To solve the drifting problem of objects caused by external disturbances under complex circumstances,a local discriminative tracking method based on superpixel is proposed.First,the ob-jects from the first ten frames of a video are segmented into superpixels,which are clustered by the k-means algorithm to construct the initial dictionary.Secondly,a linear classifier is trained by the training sample set.Then,to reduce the possibility of the object drifting,the likelihood function is defined as a linear weighted combination of the initial classifier and the updated classifier.Finally, under the particle filter framework,the particle with the highest likelihood confidence is considered as the tracking result.The dictionary and the parameters of the classifier are updated once every U frames to capture the variation of the object appearance.The simulation results show that the pro-posed algorithm can track the object under the complex circumstance with object occlusion and illu-mination change.