An Improved QPSO Algorithm Based on EXIF for Camera Self-calibration

Abstract

Binocular vision technology is an important branch of computer vision technology, which is widely used in robot motion, navigation, surgical treatment and many other fields. As is a crucial link, it is the basis of binocular vision technology to obtain the internal parameters of a digital camera. Traditional calibration methods, such as Zhengyou Zhang's method needs a calibration board, while the self-calibration method based on active vision needs to strictly control a camera to move in a designated way. Based on that, those methods can't be applied to simple and convenient occasions. In this paper, we aim to propose a new method of camera self-calibration by improving an existing QPSO algorithm with the EXIF information of digital camera photos. The method only needs to shot one object twice on different angles. We derive the conversion formula of equivalent focal length and pixel focal length and use it to initialize the algorithm. It is to find the optimal solution of the cost function transformed from the Kruppa equation by using the QPSO method. The experiment results proved that the improved method is better than the initial one and using the EXIF information to initialize the algorithm is feasible.