Towards 3D Scanning from Digital Images by Novice Users

Abstract

The uptake of hobbyist 3D printers is being held back, in part, due to the barriers associated with creating a computer model to be printed. One way of creating such a computer model is to take a 3D scan of a pre-existing object using multiple digital images of the object showing the object from different points of view. This document details one way of doing this, with particular emphasis on camera calibration: the process of estimating camera parameters for the camera that took an image. In common calibration scenarios, multiple images are used where it is assumed that the internal parameters, such as zoom and focus settings, are fixed between images and the relative placement of the camera between images needs to be estimated. This is not ideal for a novice doing 3D scanning with a “point and shoot” camera where these internal parameters may not have been held fixed between images. A common coordinate system between images with a known relationship to real-world measurements is also desirable. Additionally, in some 3D scanning scenarios that use digital images, where it is expected that a trained individual will be doing the photography and internal settings can be held constant throughout the process, the images used for doing the calibration are different from those that are used to do the object capture. A technique has been developed to overcome these shortcomings. It uses a known printed sheet of paper, called the calibration sheet, that the object to be scanned sits on so that object acquisition and camera calibration can be done from the same image. Each image is processed independently with reference to the known size of the calibration sheet so the output is automatically to scale and minor camera calibration errors with one image do not propagate and affect estimates of camera calibration parameters for other images. The calibration process developed is also one that will work where large parts of the calibration sheet are obscured.