Jean-Pierre DOYON
Point Cloud Mesh Generation and Rendering
Abstract:The general goal of this project is to develop a system able to reconstruct the 3D structure of an object, with only minimal knowledge about the camera, the object, and the relative position of the two; namely, the object is textured and rigid and assuming a perspective camera model. This system is able to acquire images from an uncalibrated, general purpose camera, extract feature points in the acquired images and track those points in the stream of images. Those points that have been tracked for a long period of time are then used to reconstruct the 3D structure of the object in the form of a 3D point cloud. The images, the tracked points for each image and the 3D reconstructed points are then used to create a realistic rendering of the object. This report will focus on the last part of the project, namely the realistic rendering of the object. It relates all the steps taken to achieve the final goal, including numerous wrong paths. The reasons for taking these paths as well as the explanations of why these were wrong are also given.