An automatic online camera calibration system for vehicular applications

Nowadays, the camera online calibration module has been a fundamental and often requested component in an Advanced Driver Assistance System (ADAS). The proposed system provides an efficient and practical solution to such request, and also shows technical advances compared to earlier systems. It utilizes the lane markings for camera orientation calibration. At each image frame, multiple vanishing points of the lane markings are estimated using the weighted least squares method, followed by a tracking process with Kalman Filter for better consistency and robustness. With the filtered vanishing point the camera extrinsic tilt and pan angles can be estimated. Even though the proposed system relies on the lane markings for calibration, unlike other similar systems, the number of the lane markings in the system is not restricted, and the shape of the lane markings is also not restricted to straight lines or any other pre-defined models. A Monte Carlo evaluation scheme is devised for that purpose using real world driving sequences. The final result has shown that given an initial calibration error, in ±4 degree interval w.r.t. ground truth for both tilt and pan angles, the proposed system is capable of converging to accurate angles and providing consistent results.