一种基于独立运动的人群计数方法

文中的工作旨在利用人体的独立运动信息,对监控视频中的人群进行计数。首先,检测当前图像中运动物体的角点特征,通过文中的特征可靠性假设,从中选取可靠的角点进行追踪,获得它们的运动轨迹。然后,利用这些特征点的空间信息进行初次聚类,限制最终聚类的数量。最后,利用角点的运动信息定义轨迹相似度和类间相似度,通过构造最小生成树的方法,进行运动一致性聚类,获得最终的特征群,从而得到运动人体的数量。

Counting people in crowds based on independent motion

The goal of this work is to utilize the information of independent motion to count people in crowds in the isurveillance video.Several steps are taking to realize that： firstly,pick out good features from all corners detected in the current image based on the hypthesis in thesis,and track them in the sequence images;secondly,the spatial distance between each pair of the locus is used to make a roughly clustering to constrain the number of final clustering hypthesis;thirdly,define the coherent motion likelyhood by the likelyhood of trajectories,merge or disjoint the prior clusters through building minimum spanning trees.One result cluster represents one person which has at least three feature points that is not in a line.Experiments of this algorithm get good results.