Predicting Unsafe Driving Behavior
<< Back to Research Areas & Projects
This work involves a novel vision-based method for bus driver fatigue detection and attention estimation for driver state monitoring. It performs head-shoulder detection, multi-pose face detection, multi-model eye detection and eye feature recognition, as well as multi-model fusion for driver state recognition. The core innovative techniques are the approach of eye openness recognition based on spectral regression and the fusion of multi-model detections for driver state estimation. It relates to human body and facial feature analysis from vision input.
We have developed video analytic tools for predicting unsafe driving behavior using face analysis (actual driving sequences cannot be shown because of anonymity and security, only simulation is shown above). This system is able to predict the mood and sentiments of the driver, like distraction, drowsiness, frequent head turnings and eye blinking (as shown in the above simulation image). These analytic tools would be deployed as monitoring devices inside SMRT buses and taxis, licensing through a local company in Singapore.
Details:
Submitted to IEEE Transaction on Intelligent Transportation Systems and also as a Technological Disclosure.