A New Method for Calibrating Gazis-Herman-Rothery Car-Following Model

摘要

Traffic simulation at the microscopic level utilizes car-following model to describe vehicle interactions on a vehicular lane. The most widely used car-following model is the Gazis-Herman-Rothery model, which contains two coefficients: m and l. The coefficients should be determined in calibration tests where the involved vehicles are tracked for their positions, velocities, and accelerations. The existing calibration methods are costly. This study proposes a calibration method using computer vision. Two computer vision algorithms are evaluated, namely, multilayer and Eigen background subtraction. The vehicle movement is tracked on a perspective plane and then is projected to an orthogonal plane. From the verification tests, we determine that the multilayer algorithm has 96.6 % accuracy for the vehicle position and 88.9 % for the velocity. The Eigen algorithm has 92.9% accuracy for the vehicle position and 84.3% for the velocity. The estimated model coefficients is 0.4 for m and 1.2 for l. These values are within the range of the most reliable coefficients according to many literatures.