Detection of Vehicle Position and Speed using Camera Calibration and Image Projection Methods

Abstract

Traffic congestion is the main problem faced by big cities, such as Jakarta. One approach to reduce congestion levels is to improve traffic management that regulates and controls the number of vehicles. To evaluate the impact of traffic management before direct implementation on the highway, traffic modeling can be carried out. Parameters in modeling traffic must be determined from a calibration process where the vehicle is accurately measured for its position and speed. This study aims to propose an efficient calibration procedure with accurate results, based on recorded vehicle movement in perspective view. First, the road image is projected using the Direct Linear Transformation (DLT) method, then the vehicle position is detected using the Background Subtraction and tracked using Mixture of Gaussian (MoG) to determine the vehicle speed. Finally, we develop a prototype of Automated Traffic Flow Monitoring based on Python programming. In the experiment results, the accuracy of vehicle position detection is evaluated based on the Euclidean distance. The average difference between the results of position detection with ground-truth is 12.07 pixels with a camera angle 40 °. The percentage of speed measurement accuracy using the DLT projection method is 96.14%.