Monitoring the Use of HOV and HOT Lanes

Abstract

This report presents the formulation and implementation of an automated computer vision and machine learning based system for estimation of the occupancy of passenger vehicles in high-occupancy vehicles and high-occupancy toll (HOV/HOT) lanes. We employ a multi-modal approach involving near-infrared images and high-resolution color video images in conjunction with strong maximum margin based classifiers such as support vector machines. We attempt to maximize the information that can be extracted from these two types of images by computing different features. Then, we build classifiers for each type of feature which are compared to determine the best feature for each imaging method. Based on the performance of the classifiers we critique the efficacy of the individual approaches as the costs involved are significantly different.