文章基本信息

标题：Regression Based Crowd Density Estimation
本地全文：下载
作者：AR.Arunachalam ; Dr. Robert Masilamani
期刊名称：International Journal of Innovative Research in Science, Engineering and Technology
印刷版ISSN：2347-6710
电子版ISSN：2319-8753
出版年度：2015
卷号：4
期号：1
页码：18691
DOI：10.15680/IJIRSET.2015.0401096
出版社：S&S Publications
摘要：We present a privacy-preserving system for estimating the size of inhomogeneous group, composed ofpedestrians that travel in different ways, without using explicit component segmentation or tracking. First, the group issegmented into objects of homogeneous motion, using the compound of dynamic textures motion model. Second, a set ofimportant feature is extracted from every segmented area, and the related features and the number of people per segment islearned with Bayesian Process regression. In this paper two Bayesian regression models are examined. The first model is acombination of Gaussian process regression with a compound kernel, which associates for both the global and local trendsof the count mapping but is limited by the real-valued outputs that do not match the discrete counts. This limitation isaddressed with a second model, which is depends on a Bayesian treatment of Poisson regression that introduces a priordistribution on the linear weights of the model. The two regression-based crowd counting methods are evaluated on a bigpedestrian data set, containing very distinct camera angles, pedestrian traffic, and odds, such as bikes or skateboarder.Research results show that regression-based counts are accurate regardless of the group size, outperforming the countcalculates produced by state-of-the-art pedestrian detectors. Finally, result of the system running on 2 h of video. Itdemonstrates the efficiency and robustness of the regression-based crowd size estimation over long periods of time.
关键词：Bayesian regression; crowd analysis; Gaussian processes; Poisson regression; surveillance.