Improving the performance of edge localization techniques through error compensation

The performance of subpixel edge localization (EL) techniques can often be improved through the compensation of the systematic portion of the localization error. In order to prove this fact, we propose and analyze a method for estimating the EL characteristic of a given subpixel localizator through statistical analysis of appropriate test images. Such estimate can be used to characterize the compensator independently of the subpixel EL technique that is being used. In order to evaluate the impact of the proposed compensation technique on the performance of a subpixel edge localizator, we have embedded it into a camera calibration procedure and compared the accuracy of the calibration with and without compensation. The improvement in the calibration precision has been proven to be significant (44%), which can be of crucial importance especially in applications of low-cost photogrammetry and 3D reconstruction from multiple views.