实时超像素跟踪算法  

Real-time Superpixels Based Tracking Method

作　　者：王暐[1] 王春平[1] 付强[1] 徐艳[1] 

机构地区：[1]军械工程学院2系,石家庄050003

出　　处：《电子与信息学报》2016年第3期571-577,共7页Journal of Electronics & Information Technology

基　　金：国家自然科学基金(61141009)~~

摘　　要：建立有效的目标表观模型是视觉跟踪算法的关键。该文采用中层次视觉线索(超像素)对目标表观进行建模,提出一种实时超像素跟踪(RSPT)算法。算法采用K近邻(KNN)方法从超像素特征集合中学习目标的判别式表观模型;在后续帧中,根据学习到的表观模型计算目标-背景置信图,然后巧妙地采用积分图方法估计目标状态,实现了高速的全局最优估计;最后设计了目标表观模型的在线更新策略,引入遮挡因子对遮挡进行判断。在配置i5处理器的电脑中,所提RSPT算法使用未经优化的Matlab代码以19帧/s的速度实时运行。对若干序列的对比实验表明,所提算法能够在多种复杂环境下稳定跟踪目标,具有良好的鲁棒性。Target appearance model is crucial for tracking. In this paper, a Real-time Super Pixels based Tracking（RSPT） method is proposed in a tracking-by-detection framework, by investigating mid-level vision cue superpixels. Firstly, a discriminative appearance model is constructed relying superpixels feature and K-Nearest Neighbor（KNN） learning method. Then the tracking problem is posed by computing a confidence map, and detecting the best target station by maximizing an object location likelihood function. The integral image data structure is adopted for fast detection, innovatively. Implemented in MATLAB without code optimization, the proposed tracker runs at 19 frames per second on an i5 laptop. Extensive experimental results on challenging sequences show that the proposed algorithm performs favorably against some state-of-the-art methods in terms of accuracy and robustness.

关 键 词：视觉跟踪 超像素 判别式表观模型 K近邻方法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]