dc.contributor.advisor |
Morris, J |
en |
dc.contributor.advisor |
Patel, N |
en |
dc.contributor.advisor |
Biglari-Abhari, M |
en |
dc.contributor.author |
Butt, Muhammad |
en |
dc.date.accessioned |
2017-01-22T23:14:17Z |
en |
dc.date.issued |
2017 |
en |
dc.identifier.uri |
http://hdl.handle.net/2292/31641 |
en |
dc.description.abstract |
Real-time object detection is critical for many computer vision applications such as surveillance and security, human activity monitoring, hazards and obstacles detection in intelligent vehicle and advanced driving assistance systems and service robots. Object detection from 2D images is very common, but it suffers from illumination changes, shadows, background clutter etc. in a dynamic environment. For a more reliable object detection system, these problems can be efficiently overcome by exploiting 3D, as an additional cue, obtained from stereo images. A number of algorithms has been developed which use 3D information as an axillary cue to detect object of interest from the scene. This thesis focuses on development of parallel contour generation leading to an object detection system with a real-time performance using real-time high resolution 3D data (depth map). A depth map of a scene is a multivalued grey image and contour generation from such an image is computationally expensive. An efficient crack code based parallel contour generation from multivalued images was developed and implemented on a parallel hardware - Graphics Processing Unit (GPU). It generates approximately 25000 contours from a 128 grey-level 3D map in < 20ms with a single scan line latency. A parallel algorithm for computation of contour moments, defining contour properties, was also developed and implemented on GPU. This parallel method for computation of contour moments is very fast and efficient. On a GTX 980 GPU, it computes moments of approximately 26000 contours, extracted from a high resolution depth map of a real scene, in < 1ms. These two lead to a fast object detection on an accelerated hardware which uses only 3D data to detect objects from a 3D scene. An improved version using previously ignored monocular points which is more robust to occlusion and merging of close objects. The improved object detection system can detect objects from a high resolution 2 Mpixel 3D scene in 30ms. |
en |
dc.publisher |
ResearchSpace@Auckland |
en |
dc.relation.ispartof |
PhD Thesis - University of Auckland |
en |
dc.relation.isreferencedby |
UoA99264922104402091 |
en |
dc.rights |
Items in ResearchSpace are protected by copyright, with all rights reserved, unless otherwise indicated. Previously published items are made available in accordance with the copyright policy of the publisher. |
en |
dc.rights.uri |
https://researchspace.auckland.ac.nz/docs/uoa-docs/rights.htm |
en |
dc.title |
Real-time 3D Contour Analysis for Object Tracking |
en |
dc.type |
Thesis |
en |
thesis.degree.discipline |
Electrical and Electronic Engineering |
en |
thesis.degree.grantor |
The University of Auckland |
en |
thesis.degree.level |
Doctoral |
en |
thesis.degree.name |
PhD |
en |
dc.rights.holder |
Copyright: The author |
en |
dc.rights.accessrights |
http://purl.org/eprint/accessRights/OpenAccess |
en |
pubs.elements-id |
609185 |
en |
pubs.org-id |
Engineering |
en |
pubs.org-id |
Department of Electrical, Computer and Software Engineering |
en |
pubs.record-created-at-source-date |
2017-01-23 |
en |
dc.identifier.wikidata |
Q112932035 |
|