期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2010
卷号：XXXVIII - Part 5
页码：207-212
出版社：Copernicus Publications
摘要：Elderly monitoring through video surveillance systems may be a solution for detecting humans' falls. Traumas resulting from falls in the third age have been reported as the second most common cause of death, especially for the elderly. However, such monitoring architectures are usually too costly due to the high human resources required. Furthermore, they violate privacy. For this reason, a strong research interest is recently focusing on the use of advanced computer vision tools for detecting falls of elderly people. In this paper, we propose a real-time computer vision–based system capable of automatically detecting falls of elderly persons in rooms, using a single camera of low cost and thus low resolution. The system is automatically reconfigured to any background changes which cover not only light modifications but also texture alterations. In addition, the proposed method is not sensitive to camera movement, permitting the use of active cameras in person monitoring. Another important innovation of the proposed scheme is its independence from the direction of the fall, in contrast to reported approaches in which fall is usual ly identified only as rapid vertical motion. To achieve these goals, the system initially extracts the foreground object from the background using a background subtraction approach. Dynamic background modelling is achieved using an iterative approach which actively updates parts of the background texture, according to the level of motion in the scene and the location of the moving object. Regarding estimation of motion information, our approach exploits motion vectors information which is directly available from the MPEG coded video streams. This information is projects onto "good" descriptors to improve system reliability
关键词：Humans' fall detection; dynamic background modelling; visual analysis; motion estimation
