Stereo Camera Re-Calibration and the Impact of Pixel Location Uncertainty

摘要

This report presents an algorithm for the re-calibration of a stereo camera system mounted on an autonomous vehicle. The algorithm employs feature- based matching in combination with an iterative approach to estimating the fundamental matrix with the RANdom SAmple Consensus (RANSAC) paradigm followed by a constrained nonlinear estimate. Simulations indicate that stereo camera re- calibration is possible with the algorithm and the accuracy of the process depends on the amount of uncertainty in the pixel location of corresponding left and right image points. Future work will be directed at improving the accuracy of determining corresponding points.