Motion estimation on driver video data

In this paper, we analyze driver behaviour from video data of the driver taken as he or she drives the vehicle. An optical flow-based motion estimation is used to determine when the driver changes pose. The proposed motion analysis is tested on a video sequence of an actual driver. The results indicate that the motion estimation can, with probability 80%, correctly detect pose transitions. Experimental results are given which relate the probability of correct detection to the number of false detections.