Abstract for HONS 01/20
Utilizing 3D scans for Labelling and Visual Recognition
Timothy Chang
Department of Computer Science and Software Engineering
University of Canterbury
Abstract
Labelling of 3D scans is an important task in computer vision. There are many algorithms that have been developed to perform visual recognition on 3D scans. Labelling 3D scans is often the first step in producing more advanced visual recognition algorithms. This reports details a method to label large 3D scans. It uses region growing, k-nearest neighbours, normal estimation, color space transformation and human input to produce instance level labels for use in machine learning. This report evaluates this method on the Stanford 3D Indoor Spaces Benchmark and compares it's performance with modern supervised learning techniques.