摘要
Mean Shift行人跟踪采用颜色特征直方图作为跟踪特征,存在易受背景颜色干扰等问题。基于此,在传统的Mean Shift行人跟踪算法中引入粒计算的思想,提出粒化的Mean Shift行人跟踪算法,对图像目标区域作粒层分块来提取块颜色特征信息,并在颜色特征表示上作不同粒度的粒化,最后在Mean Shift迭代框架下实现行人跟踪。该方法相比传统的跟踪方法具有计算复杂度更低、稳健性更好的优点。在PETS2009和CAVIAR数据库上的实验表明,这种方法跟踪正确率更高,在颜色干扰下稳健性更好,能够实时有效地跟踪行人。
Mean shift pedestrian tracking that uses a color histogram as its tracking feature has drawbacks, e. g.,performance can easily be affected by the introduction of a background color. To solve this problem, the idea ofgranular computing was introduced into the traditional mean shift pedestrian tracking algorithm, and a new granularmean shift pedestrian tracking algorithm, based on granular computing, is presented. The algorithm blocks the image, s target area with specific granularity to extract color features, then adopts different color channels of granulationon the feature, and finally realizes target tracking under the framework of the mean shift iteration. Comparedwith other traditional methods the algorithm displays lower computational complexity and is more robust. Experimentalresults on PETS2009 and CAVIAR databases show that the algorithm achieves a higher tracking accuracy, betterrobustness and efficiency under color interference, and can track the target pedestrian in real time.
出处
《智能系统学报》
CSCD
北大核心
2016年第4期433-441,共9页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金项目(61273304)
上海市中医药三年行动计划重点项目(ZY3-CCCX-3-6002)
关键词
信息粒
粒计算
MEANSHIFT
特征提取
行人跟踪
information granules
granular computing
mean shift
feature extraction
pedestrian tracking
