Camera calibration without feature extraction

We present an original approach to the problem of camera calibration. Contrary to classical techniques, which first extract the image features and then compute the camera parameters, we directly search for the camera parameters that best map three-dimensional points onto the image edges, characterized as maxims of the intensity gradient or zero-crossings of the Laplacian. Expressed as a one-stage optimization problem over the parameters of the camera, the whole calibration process is solved by classical iterative optimization. We describe experiments on synthetic and real data.