A closed-loop crash warning system based on heterogeneous cues

The area of human machine interaction has been immersed into transportation research for many years and has embraced in intelligent transportation systems for the development of next-generation active safety systems in recent years. It has long been identified that the driver distraction plays a major role in traffic accidents and application of imaging technologies to detect and predict such critical situations has become attractive. This paper presents a closed-loop image based solution to detect critical driver behaviors and produce warning signals when safety critical situations arise. The system tracks driver upper body, eye movements and external objects using multiple cameras in a vehicle to determine the driving risk. Current implementation detects three risky driving behaviors, namely, texting, drinking and reach to grab objects while driving, and combines them with gaze movement to determine the crash severity.