摘要
针对光流法在估计人群运动速度时对噪声敏感的问题,提出一种基于互信息(Mutual Information,MI)纠正的人群运动特征提取方法。以稠密光流场为数据空间,在局部可重叠区域内计算各个速度向量与区域速度期望之间的无参数测度MI,通过直方图统计出MI偏小的速度向量,然后将这些向量向区域速度期望向量做出纠正处理,直至MI符合预设条件为止。实验表明,经MI纠正后可得到光滑的速度场,为后续分析人群行为提供增强的人群运动特征。
Traditional optical flow method is highly sensitive to noise when estimating crowd velocity field. This paper puts forward a new crowd speed computing approach based on mutual information(MI) rectifying algorithm. The approach calculates dense optical flow field as base data space in which local overlapped rectangle regions are separated, and MIs in a rectangle are achieved between each veloci- ty vector and the mean vector. Then the lower MIs are obtained by employing the histogram method and all of them are rectified in terms of the mean vector belong to the rectangle until they satisfies preconditions. The experimental results show that the paper obtains smooth optical flow via the method so as to provide enhanced crowd motion features for the following crowd behavior analysis work.
出处
《智能计算机与应用》
2012年第4期1-3,共3页
Intelligent Computer and Applications
基金
国家自然科学基金资助项目(61171184)
黑龙江省自然科学基金资助项目(F201021)
关键词
互信息
光流法
人群运动
特征提取
Mutual Information
Optical Flow Method
Crowd Motion
Feature Extraction