Comparison between eigenfaces and Fisherfaces for estimating driver pose

Abstract

In this paper, we discuss the problem of estimating the pose of an automobile driver from video of the driver as he or she drives the vehicle. The results reported are a follow-on to those presented in the IEEE Intelligent Transportation Systems Conference 2000 by the same authors. The previous results pertained to pose classification using a non-parametric eigenface approach. Although the eigenface approach yielded impressive results, there were certain types of mis-classification errors that could be eliminated perhaps by using a different approach. In this paper, classification results obtained by another non-parametric approach, namely Fisherfaces, are compared with the eigenface approach. These results show that Fisherfaces outperform eigenfaces.