A stratified optimization method for camera self-calibration

We present an efficient stratified optimization approach for self-calibration of a camera in the case that its focal length and the principal point location are unknown. Generally we can assume that the two views are of the same focal length, and the pixels are nearly perfectly rectangular, also it is possible to know the aspect ratio rather accurately. In our approach, we use singular value decomposition to solve a modified Kruppa Equation to derive the focal length with the supposition that the principal point is at the center of the image, and perform an exhaustive search for the principal point near the center of the image to minimize a cost function. We can get a much accurate result with the optimized principal point location.