Optimal Vanishing Point Estimation with Performance Evaluation

A method for automatically and optimally determining the vanishing points from a single image is proposed in this paper. This method is based on the line segments and a random sample consensus (RANSAC) algorithm, leading to maximum likelihood estimation (MLE) results. The detection is performed without any prior knowledge of the camera parameters and additional properties of underlying 3D lines. An analysis of the error propagation is done in order to give quantitative elements on the precision of the detected vanishing points. The vanishing points detection method is experimentally validated on a set of 100 images.