Real-time pose classification for driver monitoring

Abstract

Driver pose estimation is one of the key components for future driver assistance systems since driver pose contains much information about his driving condition such as attention and fatigue levels. To this goal, a system is presented that detects the pose of the driver face in real time under realistic lighting conditions. The goal of the work is to automate the training phase, thereby eliminating the process of entering user information as much as possible. Two learning methods are presented for driver pose estimation. The first method uses unsupervised learning with Kohonen competitive networks, while the second method explores SVR with an appearance-based method.