Focal Length Changes Estimation on Zooming Stereo Camera using Fundamental Matrix

Abstract

Fundamental matrix is algebraic representation of relation between two images taken from different camera position and orientation. It is often used to reconstruct 3D objects from images taken by uncalibrated cameras. On uncalibrated cameras, we have no information of focal length thus the reconstructed object is not matched to real world units. If the images are zoomed, the corresponding points in images are scaled and affect the result of reconstructed object. To avoid it, the focal length values of each image is required. During zooming process, it's impractical to do manual calibration of each frame. Self calibration could be used but it requires certain knowledge of the scene such as 3D parallel lines on the scene. This make difficult to know the focal length of image plane if the recorded scene is completely unknown. In this paper, we did focal length change estimation using information within fundamental matrix of stereo images. With this method we can reduce the process of point matching needed to reconstruct three dimensional objects. In order to solve the problem we tried to exploit the elements of fundamental matrix of stereo image pairs to obtain the estimate the value of focal length by using intrinsic parameters of minimal zoom as reference. In this paper, we managed to use fundamental matrix to estimate the changes of focal length relative to known focal length. The estimation result shows there is slight difference compared to estimation using points distance.