Contributions to 3d reconstruction using passive and active methods

Abstract

Reconstructing the three dimensional structure of an object or a scene from the two dimensional images captured using different sensors is a significant problem in computer vision. Several techniques have been developed to recover the geometric structure of real world objects and is applied in medical imaging, inspection of mechanical parts, object recognition, replication of cultural heritage works, gaming, etc. The performance of 3D reconstruction methods depends mainly on the accuracy of reconstruction, computation complexity and speed. In the journey towards the development of an efficient 3D reconstruction method, passive and active 3D reconstruction methods are investigated and some of the well-known methods are chosen based on the information available in the existing literature. These methods are modified to improve the performance of 3D reconstruction. The objective of this research is to propose methods for reconstructing the 3D structure of real world object or scene using the images captured using different sensors and compare the performance of these methods based on accuracy, complexity, and speed. Obtaining the accurate depth information using a single image is difficult and thus there is a need to use two or more images for performing 3D reconstruction. In the proposed research, a passive method of 3D reconstruction is performed using stereo vision. Two images of the scene or an object is used to obtain the depth information. Keypoints are extracted using Scale Invariant Feature Transform (SIFT) and matched. The number of keypoints obtained using SIFT is more when compared to the features obtained using other feature extraction methods and hence it is used. Camera calibration is performed using the camera calibration toolbox. The fundamental matrix is calculated and epipolar rectification is done. The disparity map is calculated using region based stereo algorithm using global error energy minimization. Finally, the depth map is obtained from the disparity map. The results obtained are compared with ground truth values and is observed that the depth information is not recovered in few areas. To overcome this drawback, multiple images are used for performing the 3D reconstruction. newline