Modeling Impact of Sensor Placement for Vision-Based Traffic Monitoring

Abstract

A generic sensor placement model for vision-based traffic monitoring is the focus of this study. A significant problem with such sensors is the difficulty in detection because of the occlusion between vehicles. Thus, the efficiency of traffic monitoring can be directly affected by sensor placement. To simulate various traffic flows, models are developed for various aspects of moving traffic. Such models include Gaussian mixture distributions for vehicle dimensions and the distribution for gap length between vehicles. These models are used to predict the vehicle detection error in a traffic flow as perceived from various sensor locations when vehicle headlight detection methods are used. Validation of the model has shown that accuracy is consistent with performance from a vehicle detection framework with approximately 3% variance on average.