Correcting radial lens distortion with advanced outlier elimination

Radial lens distortion occurs when low-cost cameras or cameras with wide-angle or fish-eye lens are used. The line-based approach is based on the property that straight lines in the 3D world must be straight in 2D images; it estimates a radial lens distortion parameter by finding curved lines in the 2D image which are straight lines in 3D world. To estimate the parameter more accurately, a conventional algorithm eliminates outliers, which are defined as lines which are curved both in the 2D image and in the 3D world. We propose a novel algorithm which extends the concept of outlier to include lines which can cause errors in the estimate of the parameter even though the lines are straight in the 3D world. The proposed algorithm can eliminate outliers effectively. As a result, the proposed algorithm can estimate the distortion parameter more accurately than existing algorithms.