Artificial Intelligence based Camera Calibration

Abstract

Camera calibration technique plays a vital role in three dimensional computer vision systems. The aim of this technique is to calibrate the camera in order to collect more precise three dimensional data from the images to utilize for robot navigation, three dimensional reconstruction, biomedical, virtual reality and visual surveillance. In camera calibration one of the major issue is to search out the set of image parameters describing the mapping between three dimensional images reference coordinates and two-dimensional images reference coordinates. Currently, MATLAB toolbox and OpenCv are the most popular tools used by the researchers for camera calibration. We utilize the concept of deep learning to recognize the chessboard corners. Our presented technique is a convolutional neural network (CNN) trained on a huge number of chessboard images. The network is trained on different datasets: noisy and with high lens malformation images. The proposed scheme is more accurate than the conventional MATLAB algorithm technique. The presented technique is more accurate against the different sort of ruination present in the training set. Results reaffirmed the correctness and effectiveness of our proposed CNN technique.