Monocular 3D Head Tracking to Detect Falls of Elderly People

Abstract

Faced with the growing population of seniors, Western societies need to think about new technologies to ensure the safety of elderly people at home. Computer vision provides a good solution for healthcare systems because it allows a specific analysis of people behavior. Moreover, a system based on video surveillance is particularly well adapted to detect falls. We present a new method to detect falls using a single camera. Our approach is based on the 3D trajectory of the head, which allows us to distinguish falls from normal activities using 3D velocities.

3D Head Trajectory Using a Single Camera

“…A head ellipse was also tracked by Nummiaro et al [9] using a color-based particle filter. In our previous work [12], we also have used this idea to compute the 3D head localization from a 2D head ellipse tracked in the image plane. The 3D head pose was computed using a calibrated camera, a 3D model of the head and its projection in the image plane.…”

“…For instance, in our research on fall detection [12], the 3D head trajectory of a person in a room is needed. Indeed, Wu [16] showed in a biomechanical study with wearable markers that the 3D head velocities were efficient to detect falls.…”

A novel wearable smart button system for fall detection

“…The former is designed to capture the target image through a camera and then determines whether the target posture is fall behavior or not via image processing algorithms. For example, Rougier et al once put forward a fall detection algorithm through capturing the body changes, but the limitation is that such measurements could offend one's privacy and is only suitable for detection in a small area (Rougier, C et al 2006). Chen adopted three frame differential method to extract the outline of the target so as to estimate occurrence of fall.…”

Fall detection in a smart room by using a fuzzy one class support vector machine and imperfect training data

“…There are various ways to detect a fall event using computer vision and signal processing techniques. In [2] and [3], C. Rougier et al use a threshold-based algorithm to compare the values of the extracted features with the corresponding thresholds to make decisions. The head's 3-D velocity and human shape information are extracted as features respectively.…”