Automatic camera calibration by landmarks on rigid objects

Traffic Camera Calibration via Vehicle Vanishing Point Detection

“…Our experiments show that our method is not competitive against more specialized methods which utilize the assumption of a straight road segment. However when considering the parking lot scenario where these methods are not applicable and our method achieves accuracy close to the existing state of the art approach [2] while not being limited by requiring vehicles of specific make and model to be present in the scene.…”

“…We evaluate our method on two traffic surveillance datasets BrnoCompSpeed [16] and BrnoCarPark [2]. Both datasets contain information about measured distances between pairs of points on the road plane along with their pixel positions in the images.…”

“…Bartl et al [1] use iterative calculation of the focal length to minimize the weighted reprojection error of the keypoints followed by a least squares estimation of the ground plane. In a very similar approach [2] the authors employ differential evolution to obtain all of the camera parameters in single optimization scheme. These methods require that few specific models of vehicles are present in the scene, which may prevent their use in regions with different vehicle model composition or in areas with low traffic density.…”

“…The limitation to straight road segments makes the methods not applicable in various important traffic scenarios such as curved roads, intersections, roundabouts and parking lots. Methods not requiring straight road segments also exist, but they come with specific drawbacks such as requiring parallel curves to be visible on the road plane [6] or vehicles of specific make and model to be present in the scene [1,2,4].…”

“…We evaluate our method on two different datasets of traffic surveillance videos of straight roads [16] and parking lots [2]. Our experiments show that our method is not competitive against more specialized methods which utilize the assumption of a straight road segment.…”