eTheses Repository

A highly adaptable model based – method for colour image interpretation

Braik, Malik Shehadeh (2016)
Ph.D. thesis, University of Birmingham.

Loading
PDF (3621Kb)Accepted Version

Abstract

This Thesis presents a model-based interpretation of images that can vary greatly in appearance. Rather than seek characteristic landmarks to model objects we sample points at regular intervals on the boundary to model objects with a smooth boundary. A statistical model of form in the exponent domain of an extended superellipse is created using sampled points and appearance by sampling inside objects.
A colour Maximum Likelihood Ratio criterion (MLR) was used to detect cues to the location of potential pedestrians. The adaptability and specificity of this cue detector was evaluated using over 700 images. A True Positive Rate (TPR) of 0.95 and a False Positive Rate (FPR) of 0.20 were obtained. To detect objects with axes at various orientations a variant method using an interpolated colour MLR has been developed. This had a TPR of 0.94 and an FPR of 0.21 when tested over 700 images of pedestrians.
Interpretation was evaluated using over 220 video sequences (640 x 480 pixels per frame) and 1000 images of people alone and people associated with other objects. The objective was not so much to evaluate pedestrian detection but the precision and reliability of object delineation. More than 94% of pedestrians were correctly interpreted.

Type of Work:Ph.D. thesis.
Supervisor(s):Pycock, David
School/Faculty:Colleges (2008 onwards) > College of Engineering & Physical Sciences
Department:School of Engineering
Subjects:TK Electrical engineering. Electronics Nuclear engineering
Institution:University of Birmingham
ID Code:6927
This unpublished thesis/dissertation is copyright of the author and/or third parties. The intellectual property rights of the author or third parties in respect of this work are as defined by The Copyright Designs and Patents Act 1988 or as modified by any successor legislation. Any use made of information contained in this thesis/dissertation must be in accordance with that legislation and must be properly acknowledged. Further distribution or reproduction in any format is prohibited without the permission of the copyright holder.
Export Reference As : ASCII + BibTeX + Dublin Core + EndNote + HTML + METS + MODS + OpenURL Object + Reference Manager + Refer + RefWorks
Share this item :
QR Code for this page

Repository Staff Only: item control page