基于字典学习的粒子滤波鲁棒跟踪算法

字典学习广泛应用于图像去噪、图像分类等领域，但是将离线字典训练如何应用于视频目标跟踪的研究较少。本文采用一种字典编码方法提取目标的局部区域描述符，通过训练分类器将跟踪问题转化为背景和前景二值分类问题，并通过粒子滤波对物体位置进行估计实现跟踪。不同图像序列的实验结果表明，与现有的方法相比本文的算法具有较好的鲁棒性。

Object Tracking Based On Dictionary Learning Joint Practical Filter

The dictionary learning is widely used in image enhancement, image classification and so on. In field about object tracking, dictionary learned offline is little used. In this paper , we represent local image patches of a target object by codes with an dictionary learning offline. We pose object tracking as a binary classification problem by a learnand classifer and a practical filter is used to estimate the tracking result. We do the simulations on a variety of challenging sequences . Experiments on different sequences with evalution of the state-of-the art methods show our algorithm performs favorable performance.