3D Cloud Reconstruction Using Digital Images from Two Viewing Angles

In this paper, a 3D reconstruction technology using Direct Linear Transformation (DLT) algorithm is introduced to estimate the cloud shape, height and range based on digital images from two viewing angles. The quality of the reconstruction depends on the accuracy of both the projection matrix from each camera and the corresponding points matching between two images. The projection matrix is parameterized by intrinsic and extrinsic parameters. The intrinsic parameters are obtained from the camera’s specifications and extrinsic parameters, such as orientation and location can be measured directly using a digital protractor, compass and GPS. Correspondence matching is to select the same features in the scene from both images, and an automatic correlation-based method is proposed in this paper to identify and match points along cloud edge. The proposed matching algorithm is first to detect the cloud edges in both images based on Minimum Cross Entropy (MCE) threshold, and then the points along cloud edges are chosen as the feature points/matching candidates, and subsequently cross correlation based method is used to find the best matches. The matching process can be completed more efficiently comparing to an exhaustive search, since the matching candidates are along the edge and the number of matching candidates is significantly reduced. Simulations are developed to test and verify the reconstruction. Preliminary results show that given reasonable separation of two cameras and object range, we can obtain a reasonably good 3D representation of the object. Using the same tool, we can evaluate the expected accuracy of 3D reconstruction for field experiments. Reconstructions of multiple clouds are provided in this paper to demonstrate the feasibility and efficacy of the proposed technique.