基于SURF的高密度人群计数方法  

High-Density Crowd Counting Method Based on SURF Feature

作　　者：梁荣华[1] 刘向东[1] 马祥音[1] 王子仁[1] 宋明黎[2] 

机构地区：[1]浙江工业大学计算机科学技术学院,杭州310023 [2]浙江大学计算机科学与技术学院,杭州310027

出　　处：《计算机辅助设计与图形学学报》2012年第12期1568-1575,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61070114;61202205);浙江省自然科学基金重点项目(Z1090630);浙江省自然科学基金(LY12F02037);浙江省公益性技术应用研究计划项目(2012C23122)

摘　　要：为了解决在高密度人流或视场开阔环境下人群计数准确率低的问题,提出一种基于SURF的高密度人群计数方法.首先采用最小生成树改进了传统的基于密度的聚类算法,使其最小搜索域自适应聚类数据的分布;在此基础上实现运动人群的SURF特征点分类,并以此构建运动人群的特征向量,用支持向量回归机实现了对高密度人群的数量统计.实验结果表明,该方法对高密度人群的计数有较高的准确率和鲁棒性.This paper presents a SURF-based method for high-density crowd counting, focusing on overlaying the low counting accuracy in a high-density crowd or open environment. The traditional density-based clustering algorithm （DBSCAN） by adopting minimum spanning tree （MST） is improved, making its minimal search domain adaptive to the distribution of clustering data. Then the SURF features of moving crowd through the improved DBSCAN algorithm is classified. An eigenvector which can represent the moving crowd is built on the clustering results. Finally, the number of crowd through a support vector regressor （ε-SVR） is got. The experimental results confirm that the proposed method have a high accuracy and robustness to the high-density crowd counting.

关 键 词：高密度人群计数 SURF 最小生成树 基于密度的聚类算法 支持向量回归机 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]