Centi-pixel accurate real-time inverse distortion correction

Abstract

Inverse distortion is used to create an undistorted image from a distorted image. For each pixel in the undistorted image it is required to determine which pixel in the distorted image should be used. However the process of characterizing a lens using a model such as that of Brown, yields a non-invertible mapping from the distorted domain to the undistorted domain. There are three current approaches to solving this: an approximation of the inverse distortion is derived from a low-order version of Brown's model; an initial guess for the distorted position is iteratively refined until it yields the desired undistorted pixel position; or a look-up table is generated to store the mapping. Each approach requires one to sacrifice either accuracy, memory usage or processing time. This paper shows that it is possible to have real-time, low memory, accurate inverse distortion correction. A novel method based on the re-use of left-over distortion characterization data is combined with modern numerical optimization techniques to fit a high-order version of Brown's model to characterize the inverse distortion. Experimental results show that, for thirty-two 5mm lenses exhibiting extreme barrel distortion, inverse distortion can be improved 25 fold to 0.013 pixels RMS over the image.

Removal of noise and radial lens distortion during calibration of computer vision systems

“…(2), compared to the state of literatures [13] [20] [23,24] that deal with the radial lens distortion, the proposed pattern modeling method is capable of removing the radial distortion with noise as a whole while the state of art methods have to consider the effect of the noise during correcting the radial distortions.…”

“…In [23], inverse distortion patterns different from [13] are used to correct the radial distortion. In [24], Hough transformation was used to correct the radial distortions. Unfortunately, none the above mentioned methods could remove the radial distortions completely.…”

“…The most commonly encountered lens distortions are radial distortions [22] that are addressed in [13,20,23,24] with different methods. In [13], the radial distortion of the projector is corrected independently by adjusting the coefficients of the projected pattern with inverse distortion.…”

Quality Assessment of Self-Calibration with Distortion Estimation for Grid Point Images

“…The work * Corresponding author by (Devernay and Faugeras, 2001) compares a self-calibration technique against a regular calibration, but they do not use a minimal solver that models radial distortion. In (De Villiers et al, 2008), different distortion models and numerical optimization techniques are compared, but only the traditional calibration technique is used in any of the experiments.…”

OpenMVG: Open Multiple View Geometry

“…OpenMVG provides an abstract camera interface that can be used seamlessly along the library with the following concrete implementations: pure pinhole [6], pinhole with 1 to 3 radial distortion coefficients [31], pinhole with 5 distortion coefficients (3 radial + 2 tangential) (aka. Brown-Conrady) [29,30], and fish-eye [32].…”