Camera Calibration for 3D Leaf-Image Reconstruction using Singular Value Decomposition

Abstract

Features of leaves can be more precisely captured using 3D imaging. A 3D leaf image is reconstructed using two 2D images taken using stereo cameras. Reconstructing 3D from 2D images is not straightforward. One of the important steps to improve accuracy is to perform camera calibration correctly. By calibrating camera precisely, it is possible to project measurement of distances in real world to the image plane. To maintain the accuracy of the reconstruction, the camera must also use correct parameter settings. This paper aims at designing a method to calibrate a camera to obtain its parameters and then using the method in the reconstruction of 3D images. Camera calibration is performed using region-based correlation methods. There are several steps necessary to follow. First, the world coordinate and the 2D image coordinate are measured. Extraction of intrinsic and extrinsic camera parameters are then performed using singular value decomposition. Using the available disparity image and the parameters obtained through camera calibration, 3D leafimage reconstruction can finally be performed. Furthermore, the results of the experimental depth-map reconstruction using the intrinsic parameters of the camera show a rough surface, so that a smoothing process is necessary to improve the depth map.