融合显著度时空上下文的超像素跟踪算法  

Super Pixel Tracking Algorithm via Fusing Saliency Detection into Spatio-Temporal Context

作　　者：郭春梅[1] 陈恳[1] 李萌[1] 李斐[1] 

机构地区：[1]宁波大学信息科学与工程学院,宁波315000

出　　处：《模式识别与人工智能》2017年第8期728-739,共12页Pattern Recognition and Artificial Intelligence

基　　金：国家自然科学基金项目(No.60972063);宁波市自然科学基金项目(No.2014A610065);宁波大学科研基金(理)/学科项目(No.XKXL1308)资助~~

摘　　要：为了更有效地利用目标的特征信息,提高目标的跟踪精度和鲁棒性,提出融合显著度时空上下文的超像素跟踪算法.首先对目标上下文区域进行超像素分割,根据运动信息计算目标上下文的运动相关性及特征协方差信息,得到相关性显著度.然后基于贝叶斯框架,在频域构建融合显著度信息的时空上下文模型.再利用联合颜色和纹理的直方图信息计算巴氏系数,更新时空上下文模型.此外,引入尺度金字塔模型,准确估计目标尺度.最后加入低通滤波自适应运动预测模块,在线更新动态模型样本集,使用岭回归方法实现低通滤波的参数在线更新.在公共数据上的实验表明,文中算法在光照变化、背景复杂、目标旋转、机动性高、分辨率低等情况下具有较好的跟踪效果.To achieve more efficient utilization of image feature information and improve the tracking accuracy as well as robustness of the target, an improved super pixel tracking algorithm via fusing salient region detection into spatio-temporal context is proposed. Firstly, super pixe] segmentation is conducted in the context region of the target, then the motion relevance of target context and the regional covariance information are utilized to calculate the correlation saliency of the image super pixels. Based on Bayesian framework, the model fusing saliency detection into spatio-temporal context is built in the frequency domain. Next, the color and texture histograms of current frame and reference template are employed to calculate the Bhattacharyya coefficient and update the spatial and temporal context model. The scale pyramid model is introduced to estimate the target scale. Finally, the adaptive motion prediction module is incorporated by updating online dynamic model sample set and using ridge regression method todetermine the parameters of a low pass filter. Experimental results on public superiority of the proposed algorithm over other algorithms in illumination change, object rotation, high mobility, low resolution. database indicate the complex background,

关 键 词：目标跟踪 超像素 显著度 岭回归 时空上下文 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]