Automated video surveillance for preventing suicide attempts

Inmate suicide by hanging is documented as a major cause of death in prisons. Important efforts have been made to develop technological prevention tools, but the proposed solutions are mostly using cumbersome devices, in addition to their lack of generalizability. Nowadays, computer vision methods for real-time video analysis have experienced impressive progress. The recent emergence of RGB-D cameras clearly illustrates the achieved advances by offering new ways for machines to interpret human activity. There was however no significant works on exploiting this evolution, and as a result, CCTV systems used for monitoring suicidal inmates are still greatly depending on human attention and intervention. This paper proposes an intelligent video surveillance system for automated detection of suicide attempts by hanging. The proposed algorithm is able to efficiently model suicidal behavior by exploiting the depth information captured by an RGB-D camera. Activity detection is then performed by classifying visual features characterizing body joint movements. Our method demonstrated a high robustness on a challenging dataset including video sequences where suicide attempts are simulated.