An accurate and practical calibration method for roadside camera using two vanishing points

Calibrating roadside camera is essential and indispensable for intelligent traffic surveillance systems. Due to the characteristics of the traffic scenarios, the commonly used camera calibration methods based on calibration patterns are no longer suitable, since there are generally no calibration patterns (e.g., checkerboard) in traffic scenarios. In this paper, we propose an accurate and practical calibration method for roadside camera, where the vanishing point in the direction of traffic lane and the vertical vanishing point are employed that can be easily obtained from most traffic scenarios. By making full use of video information, the multiple observations of two vanishing points are available. Using these observations, we present a dynamic calibration method to correct camera parameters, where the least squares estimation is employed instead of closed-form computation. The experimental results on synthetic data show the accuracy and robustness to the noise of vanishing points; the experimental results on real traffic images demonstrate the effectiveness and practicability of the proposed calibration method.