A fall detection algorithm for indoor video sequences captured by fish-eye camera

In this paper we present an algorithm that can discriminate between standing and fallen silhouettes in video sequences acquired by a fish-eye camera, in order to detect falls in an indoor environment. The proposed algorithm exploits the model of image formation that is based on the spherical projection to derive the orientation in the image of elongated vertical structures. The algorithm does not require the camera to be calibrated. The only requirement is that the optical axis of the camera being parallel to the vertical axis. Initial results show that fall detection can be performed with high accuracy, whereas, the algorithm itself is very efficient, allowing real time implementation.