3D Reconstruction in Real Time

The project aims to achieve 3D reconstruction in real time by using multiple 2D images from the scene but with different views. The project combines machine vision, multiprocessor coding, and computer graphics to be able to acheive accurate results in real time. Initially, the camera's internal parameters (focal length, pixel size, etc...) and external parameters (location in the relation to other cameras) needed to be determined. After calibrating the cameras, these measurements are used to determine the epipolar geometry of groups of camera, so that these images can be rectified. Recitification helps with the main problem that these project looks to optimize, finding corresponding points between signals. Corresponding points are found by sliding a window across the image and comparing the normalization of all the points on the same location. Finally these points will be triangulated and their depth can be easily found. The speed of this algorithm will determine if these objects can be reconstructed in real time.