Automatic Traffic Camera Calibration Using 3D Scene Reconstruction with Structure-from-Motion and Custom Image Data

Traffic cameras play a crucial role in monitoring and managing transportation systems, and accurate calibration is essential for reliable data acquisition. Traditional calibration methods often require manual interventions, physical markers, or specific tools, which can be both time-consuming and costly, especially in complex urban environments. To overcome these challenges, recent methods use the 3D reconstruction of the scene using the Structure-from-Motion (SfM) techniques. This allows for a precise traffic camera localization in the model and then finding the intrinsic and extrinsic camera parameters. However, these methods typically rely on high-quality image data from public sources such as Google Streeve View (GSV), which does not always provide images that are up to date or sometimes are not available for a given location. In this paper, we introduce a new approach of automatic camera calibration via the SfM techniques using our own data. Testing was conducted using a custom dataset consisting of images taken from various angles, followed by the 3D reconstruction and calibration process. We evaluated our approach on the task of distance measurements in the view of the traffic camera. The results demonstrate an accuracy of measurements with an error margin of 1.03 meter, showcasing the effectiveness of the proposed method in real world applications.