Fall detection using body volume recontruction and vertical repartition analysis

Authors:
Edouard Auvinet;Franck Multon;Alain St-Arnaud;Jacqueline Rousseau;Jean Meunier
Affiliations:
University of Montreal, IGB, Montreal, QC, Canada and University of Rennes 2, M2S, Rennes, France;University of Rennes 2, M2S, Rennes, France;Health and social service center Lucille-Teasdale, Montreal, QC, Canada;CRIUGM, University of Montreal, Montreal, QC, Canada;University of Montreal, IGB, Montreal, QC, Canada
Venue:
ICISP'10 Proceedings of the 4th international conference on Image and signal processing
Year:
2010

Citing 4
Cited 0

The Visual Hull Concept for Silhouette-Based Image Understanding

IEEE Transactions on Pattern Analysis and Machine Intelligence
A Four-step Camera Calibration Procedure with Implicit Image Correction

CVPR '97 Proceedings of the 1997 Conference on Computer Vision and Pattern Recognition (CVPR '97)
Fall Detection from Human Shape and Motion History Using Video Surveillance

AINAW '07 Proceedings of the 21st International Conference on Advanced Information Networking and Applications Workshops - Volume 02
Compensated Visual Hull for Defective Segmentation and Occlusion

ICAT '07 Proceedings of the 17th International Conference on Artificial Reality and Telexistence

Quantified Score

Hi-index	0.00

Visualization

Abstract

With the life expectancy increase, more and more elderly people risk to fall at home. In order to help them living safely at home by reducing the eventuality of unrescued fall, autonomous systems are developped. In this paper, we propose a new method to detect falls at home, based on a multiple cameras network for reconstructing the 3D shape of people. Fall events are detected by analyzing the volume distribution along the vertical axis, and an alarm is triggered when the major part of this distribution is abnormaly near the floor which implies that a person has fallen on the floor. This method is evaluated regarding the number of cameras (from 3 to 8) with 22 fall scenarios. Results show 96% of correct detections with 3 cameras and above 99% with 4 cameras and more.