High-accuracy geometric camera calibration
Publication from Digital
In this paper we describe a method for a camera calibration to satisfy the ever-increasing demands for high accuracy, which is crucial to many vision- applications. High accuracy calibration results are a requirement for good results. To achieve this accuracy, we try to compensate for structural deciencies in the pattern, like creases or wrinkles, by optimizing a structural model. Of course, this method increases the time necessary to calibrate the camera, but in most cases cameras are calibrated only once. Therefore, increased processing- time is accepted as trade-o for improved accuracy. In addition, we also rene the corner or ring centers with pattern-matching. This report begins with an introduction to the mathematical foundations
of camera models such as the pin- hole model. We then describe the involved techniques in detail, compare the results with traditional OpenCV methods, concluding with possible improvements and challenges of our approach. We use dierent patterns to study the impact on performance and accuracy. For a comprehensive benchmark, we use chessboard, ring, asymmetric circles and multi-ring targets. For simulation purposes we also include various patterns printed in dierent resolutions paired with synthetic data in printer accuracy like low-dpi setting or osets. In our tests we achieved improvements over traditional methods with a gain in accuracy in the order of a magnitude.