Human Posture Recognition by Combining Silhouette and Infrared Cast Shadows

This paper proposes a multi-infrared lights system for human posture recognition, which uses as input the combination of a body silhouette and cast shadows. More precisely, in our experimental setup, we installed 4 infrared lights in the different upper corners of a room to project shadows of a person, and a camera with an infrared filter in the ceiling to capture images. A simple electronic device is used to turn on and off each light in turn to get one cast shadow by frame. To illustrate the feasibility of this approach, we consider a simple scene where the shadows are projected directly on the ground. The features used for recognition are based on the distance transforms of the combination of the body silhouette and each cast shadow. A weighted majority vote scheme is used to decide what is the corresponding posture. Our approach was validated on a new data set captured in our laboratory, and compared with a traditional mono-camera system.