Digital video processing
Activity Summarisation and Fall Detection in a Supportive Home Environment
ICPR '04 Proceedings of the Pattern Recognition, 17th International Conference on (ICPR'04) Volume 4 - Volume 04
Monocular camera fall detection system exploiting 3d measures: a semi-supervised learning approach
ECCV'12 Proceedings of the 12th international conference on Computer Vision - Volume Part III
| Hi-index | 0.00 |
The quality of life for the ageing population is associated with the ability of the elderly people living independently. Fall is a major health hazard for the elders when they live independently. How to effectively assess, response and assist those elderly patients in trouble becomes an important research topic in medical elderly care services. This paper presents the visual fall detection subsystem developing the framework of ISISEMD project. The system is able to detect person falls by taking into consideration only camera information. The system is able to perform tracking of the person using advanced image processing and computer vision algorithms event in complex and dynamic background situations. The traditional approaches for detecting a human fall is based on the use of specialized devices, e.g., accelerometers, which is not a convenient framework, especially for persons with mild Dementia. Methods for automatic defection of person fall from camera cues uses motion information of the human object. However, using only motion information, we are not able to accurately detect a fall event. This is mainly due to the fact that a fall is encountered at different directions with respect to the camera position. In addition, motion information is a noise sensitive process. For this reason, accurate foreground object detection is required. However, foreground detection using the traditional background subtraction methods suffers from the dynamic changes of the background. To address these obstacles, we proposed in this paper a combined framework for fall alert based on joint estimation of foreground object and motion scene activity. In particular, motion information is estimated over a set of "good image pixels" to eliminate the noise sensitivity. Additionally, foreground object are extracted using frame differencing and a set of rules that express shape constraints, time continuity and the detection motion information in the scene. Experimental results on lad conditions indicate a accurate detection of a person fall.