RECONSTRUCTION OF THREE-DIMENSIONAL OBJECTS BASED ON REAL TIME STEREO VISION USING WEB CAMERA

• Main Author: TOTO WIBOWO (NIM: 23207119), AGUNG
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/12830
• Institution: Institut Teknologi Bandung

ABSTRACT:<p>The existence of physical objects in the real world represented in the form of three-dimensional. Three-dimensional object is projected with our eyes so that the form of a two-dimensional image. In the computer system, the projection in two-dimensional object known by the term image. In the development of two-dimensional image it is less, so that the required projection into the 3D model (three dimension). The process of projection of the real world into the 3D model is called the 3D reconstruction. Tool called the project a 3D scanner.<p>Projection of the real world become image by remove depth information (z coordinates in the system of coordinates x, y, z). This depth of information we need in the 3D reconstruction process. To restore this information can be made with a variety of approaches one of which is stereo vision. Stereo vision is senses system with two image of the two cameras simultaneously.<p>The process of reconstruction with 3D stereo vision performed with some of the stages: stereo capture, calibration, feature segmentation, feature correspondence, reconstruction of 3D coordinates, generate 3D objects. Stereo capture is the process of using the camera with the image simultaneously. This process is done using the OpenCV library.<p>Calibration process used to find the camera parameters, such as the intrinsic parameters, and extrinsic. The process of calibration is done by using the 6x8 chessboard pattern. Parameters obtained will be used for the reconstruction of 3D coordinates. In the research, proposed calibration process with a pair of image calibration chessboard oblique. From this research proposal, to give the smallest error of 0.678.<p>Segmentation and feature correspondence is required to find the corresponding features on the image left and right. In the research proposed, correspondence by using the correspondence stereo line and stereo correspondence using the edge of the environment with the object of laser irradiation on the online space is closed. This correspondence will be compared with hybrid disparity and the results of OpenCV functions. RMSE comparison results disparitas (cover area) are: stereo edge 31.718, stereo line 17.274, opencv 64.836, while hybrid 41.551.<p>3D coordinates reconstructed from the features obtained using the triangulation formula. Collection of features that we have been reconstructed mention as a point cloud. Point cloud is then combined with the Delaunay triangulation algorithm, to be input in making the 3D object model in the form of standard 3ds files. In the research compared with the proposed reconstruction algorithm between stereo line and stereo edge correspondence. Reconstruction results using Mean Opinion Score (MOS) on stereo edge algorithm is 4.78, and stereo line algorithm is 6.48 out of 10 scale. <br />