Generic high level feature detection techniques using multi-modal spatial data

Abstract

Object and pattern recognition techniques have classically used 2-D images. Mobile-mapping systems produce images with the added modality of depth. This is motivating renewed interest in aspects of object recognition research, especially in relation to issues of scale. This thesis reports on techniques that have been developed to incorporate depth into state-of-the-art 2-D object detection and localisation methods. The techniques are empirically shown to enhance detection accuracy across a range of datasets and object types.