An Extended Method of Multiple-Camera Calibration for 3D Vehicle Tracking at Intersections

In this paper, we propose an extended method for calibration of multiple cameras for 3D vehicle tracking at intersections that eventually targets automated and accurate traffic data collection. To allow simple, efficient yet accurate camera calibration of multiple cameras, we propose to extend the method based on the traditional vanishing-point based technique for individual camera calibration. First, a common rectangular road pattern derived from parallel traffic lanes is established to set up a unique world coordinate. Then, each camera is separately calibrated using parallel lines from the road pattern and finally all cameras are jointly optimized in least-squared nonlinear approach for minimum projection error. It is evaluated in a practical traffic scene with two cameras that more than 90% accuracy is achieved for distance and vehicle 3D dimension estimation.