Abstract
A 3-D reconstruction produced using only a single camera and Structure from Motion (SfM) is always up to scale i.e. without real world dimensions. Real-world dimensions are necessary for many applications that require 3-D reconstruction since decisions are made based on the accuracy of the reconstruction and the estimated camera poses. Current solutions to the absence of scale require prior knowledge of or access to the imaged environment in order to provide absolute scale to a reconstruction. It is often necessary to obtain a 3-D reconstruction of an inaccessible or unknown enviroment. This research proposes the use of a basic SfM pipeline for 3-D reconstruction with a single camera while augmenting the camera with a depth measurement for each image by way of a laser point marker. The marker is identified in the image and projected such that its location is determined as the point with highest point density along the projection in the up to scale reconstruction. The known distance to this point provides a scale factor that can be applied to the up to scale reconstruction. The results obtained show that the proposed augmentation does provide better scale accuracy. The SfM pipeline has room for improvement especially in terms of two-view geometry and structure estimations. A proof of concept is achieved that may open the door to improved algorithms for more demanding applications.
M.Ing. (Electrical and Electronic Engineering)