People in seats counting via seat detection for meeting surveillance

CASIA OpenIR > 智能感知与计算研究中心

	People in seats counting via seat detection for meeting surveillance
	Hongyu Liang; Jinchen Wu; Kaiqi Huang
	2012
会议名称	Chinese Conference on Pattern Recognition
会议录名称	Chinese Conference on Pattern Recognition
页码	202–210
会议日期	2012
会议地点	China
摘要	People in seats counting is very important for meeting surveillance. While as a canonical pattern recognition problem, it鈥檚 very difficult due to various appearances of people and other outliers such as bags and clothes. To solve this problem we propose a coarse-to-fine framework. Firstly, we use the coarse classification module to retrieve the completely empty seats. To overcome the influence of noises caused by shadows and light spots, we fuse multiple global features calculated by background subtraction. Then in the fine classification module, a proposed SW-HOG feature and the LBP feature are combined together to solve the problem of occlusion and make sure the classification is real time. Finally a time-related calibration module is applied to suppress some outliers across frames with condition that the video frames are not successive. Experiments on a real meeting dataset demonstrate that the accuracy of the proposed method reaches 99.88%.
关键词	People In Seats countIng meeting Surveillance coarse-to-fine Classification
语种	英语
文献类型	会议论文
条目标识符	http://ir.ia.ac.cn/handle/173211/12690
专题	智能感知与计算研究中心
通讯作者	Kaiqi Huang
作者单位	中国科学院自动化研究所
第一作者单位	中国科学院自动化研究所
通讯作者单位	中国科学院自动化研究所
推荐引用方式 GB/T 7714	Hongyu Liang,Jinchen Wu,Kaiqi Huang. People in seats counting via seat detection for meeting surveillance[C],2012:202–210.