Abstract
State of the art image analysis techniques such as SURF feature detection and description are fused with muiltple view geometry theory in order to implement a Structure from Motion pipeline. The Structure from Motion pipeline tracks features in the video stream and uses the tracking data to estimate the movement of the camera. The three-dimensional structure recorded by the video is then reconstructed by using the knowledge of the cameras and the projected points to triangulate the three-dimensional locations of the points. Special considerations and modifications to existing methods are made to leverage the advantages and minimize the disadvantages of reconstructing from video versus from sets of unrelated images. The implementation is designed to take advantage of the processing power of CUDA due to the sheer amount of data involved in processing a video.