| 초록 |
In this paper, we propose an approach to improve camera calibration based on the concept of deep learning. In particular, this research focuses on enhancing the accuracy of corner detection among other camera calibration procedures. Contrary to the previous camera calibration, in order to achieve high accuracy in corner detection, a learning based algorithm is applied to corner detection, and the proposed approach reduces the re-projection error, which is one of the criteria for evaluating the accuracy of the calibration. The proposed method detects and extracts the entire checkerboard in a couple of images acquired via depth camera, and the learned and calculated weights enable the extraction and detection of the corners to be accurately carried out. In this paper, checkerboard detection and extraction improved the accuracy of corner detection by using methods, such as Prepare, FAPL, PAMFG, Reconstruction, and PADCROP. The experimental results confirmed the proposed methods, and the comparison results are also provided. Comparison focuses primarily on Zhang’s calibration algorithm, which is a commonly used calibration method. The experimental results confirmed the superiority and efficiency from a quantitative perspective in terms of the number of capture and calibration accuracy. |
