Estimating driver head pose using steerable pyramid and probabilistic learning
Nawal Alioua, Aouatif Amine, Abdelaziz Bensrhair, Mohammed Rziza · International Journal of Computational Vision and Robotics · 2015
In this paper, we propose a driver head pose estimator based on steerable pyramid transform and probabilistic learning. The steerable pyramid is used to construct a head appearance template for each considered head orientation. Then, we learn the parameters of likelihood function from a training set with a probabilistic approach. To estimate the pose of a new head image, we first apply the steerable pyramid to extract its feature vector and then the maximal value of the likelihood function computed between this vector and all pose templates are retained. We perform several tests on public Pointing '04 database to optimise the parameters of steerable pyramid, which allows to make a compromise between the accuracy and processing time. Then, we apply the optimised head pose estimator on real video sequence representing a driver in diverse attention levels. We demonstrated that our system performs a good detection of driver inattention level.