Real-time recognition of suicidal behavior using an RGB-D camera

Inmates in solitary confinement may attempt to harm themselves in many ways, resulting in trivial to mortal injuries. In this context, suicide by hanging is one of the major causes of death among the incarcerated. The Rapid detection of suicide can reduce the mortality rate. Recently, several technologies have been developed to detect suicide by hanging attempts, but most of them use bulky devices, or they are greatly depending on human attention. In this paper, we propose a computer vision based system to automatically detect suicide by hanging attempts. Our method is based on modeling suicidal actions using pose and motion features, by exploiting the body joints' positions. The proposed video surveillance system analyses depth images provided by an RGB-D camera to detect the event of interest in real-time, regardeless of illumination conditions. The experimental results obtained on a realistic dataset demonstrated the high precision of our system in detecting suicide by hanging.