The driver scene aesthetic quality is elusive and complex. The expert-based evaluation is the primary process in highway aesthetic for a long time.It is a time consuming and laborious duty. The goal of this paper is to create a statistics model which can be used to classify driver scenes into certain categories according to the distribution probabilities of the features in aesthetics. A driver scene is divided into four regions by layoutsin our study. The inherent relationship between the features of regions and experts evaluation has been studied. Based on this, three full Bayesianclassifiers using Markov Chain Monte Carlo (MCMC) algorithm and based on ordered logistic model are trained with driver scene samples. The first classifier is trained with non informative prior knowledge using large number samples. The second one is also trained with non informative prior knowledge but using fewer samples. The last model is updated posterior distribution with informative priors based on the second classifier. Deviance information criterion (DIC) is applied to ensure the suitability of every classifier. Then, three classifiers are applied to classify driver scene aesthetics and comparing it with expert-based evaluation. The result shows thatBayesian inference with appropriate informative priors can be used to classify driver scene aesthetics after updating with less number of samples. In the future, subjective evaluation of highway aesthetic could be assisted with an intelligent Bayesian classifier.
Abstract