CAR-FOLLOWING MODEL: A COMPUTER-VISION-BASED CALIBRATION METHOD

Car-following model is a mathematical model that regulates the movement of a vehicle in the longitudinal direction in microscopic level. Commonly, the model describes the vehicle movement as a function of the vehicle relative position and velocity with respect to the leading vehicle. One of the widely used models is the Gazis-Herman-Rothery model, which is characterized by coefficients m and l. The values of the coefficients vary depending on many aspects such as vehicle type and road condition. The coefficients are usually determined from a calibration test where the vehicle position, velocity, and acceleration are measured accurately. So far, a few calibration methods have been proposed; some modern methods are by using remote sensing and RTK G-PS. In the current work, the vehicle movement is recorded in a perspective view from an elevation. The recorded vehicle movement is analyzed for the vehicle position using computer vision methods. Then, the position is transformed to the actual vehicle position. Two computer vision methods are evaluated: multilayer- and Eigen-background-subtraction methods. The proposed method is evaluated to track the movement of a vehicle traveling in a short-straight distance. The results show that the tracking accuracy of the multilayer-background-subtraction method is better than the Eigen-background-subtraction method. The multilayer-background-subtraction method has 96.6% for position accuracy and 88.9% for velocity accuracy, while the Eigen-background-subtraction method has 92.9% for position accuracy and 84.3% for velocity accuracy. The most reliable car following parameters are estimated with 3.2% of error. The obtained parameter m is 0.4 and l is 1.2. Various values of m and l have been proposed by many researchers where the reliable values are in the range of 0-2.7 for m and 0-2.8 for l. Our fi ndings are within the range.