Fall Detection Using Kinect Sensor and Fall Energy Image

One of the main reasons for low acceptance by seniors the available technology for automatic fall detection is that the existing devices generate too much false alarms. Additionally, the camera-based devices do not preserve the privacy adequately. In our approach an accelerometer is utilized to indicate a potential fall. A fall hypothesis is then verified in the second stage in which we employ a depth image, which was shot at the moment of the potential fall. A detector that was trained in advance on features extracted both from depth images and points cloud is responsible for verification whether a person is lying on the floor. After all, to reliably distinguish the fall from fall-like activities we perform final verification, in which we employ the proposed fall energy image. The fall energy image expresses the distribution of the person’s motion in the set of images preceding the fall.