OBTAINING A 3D MODEL OF THE OBJECT BASED ON STEREOVISION

Abstract

A method, for obtaining a 3D model of the object with two cameras is presented. There are different methods for obtaining 3D information. The given method is based on stereovision and, in contrast to other solutions, it doesn’t need a laser, an ultrasound or any other expensive sensors. Stereovision is a technique that gives 3D information using two or more cameras. The main steps of stereovision are: calibration – the process of obtaining the camera’s internal and external parameters, rectification – the process of images alignment, stereo correspondence – the problem of getting corresponding points between images and triangulation – the process of obtaining the point’s coordinates from correspondent points and camera’s parameters. We place the object in front of the cameras and rotate it, for each rotation we have an image pair. For each image pair (left and right cameras’ images) we find correspondent points by Semi-Global Matching (SBM) algorithm. As a result, we get a disparity map. From the disparity map, having the cameras’ parameters, we obtain the depth map by triangulation and the respective point cloud – the set of points with their coordinates. Then for each point cloud we segment the object from the scene to have only the object’s point cloud. For obtaining a 3D model, we need to merge these point clouds from different scenes into one common model – this process is called registration. For registration, we use Iterative Closest Point (ICP) algorithm-sequentially merging the point clouds with the global model.