Fall detection using multi-omnidirectional cameras

Accidental falls are serious threat to life of elderly people. Even when it does not result in death, it permanantly damages physiology and psychology. Fall must be detected timely and effectively to enable early intervention. In this work we propose a method that detects falls using foreground segmentations obtained from multiple omnidirectional cameras in a Bayesian framework. We observed that the method not only successfully detects falls in videos containing different actions, but it is also robust to high noise and occlusions.