A novel robust method for camera auto-calibration in transportation surveillance

Auto-calibration is to estimate camera intrinsic parameters and extrinsic parameters from image observation. In the last decade, auto-calibration has been studied and applied in transportation surveillance. Image features from architectures, roads, vehicles and pedestrians provide the orthogonal constraints for calibration. In transportation surveillance, previous auto-calibration methods are mainly based on single-view geometry, while this paper proposes a method based on multiple-view geometry. The experiment results prove that our method is more robust to image noise than previous methods.