Browse DORAS
Browse Theses
Search
Latest Additions
Creative Commons License
Except where otherwise noted, content on this site is licensed for use under a:

Speeding up Adaboost object detection with motion segmentation and Haar feature acceleration

Ramachandruni, Radha Krishna (2009) Speeding up Adaboost object detection with motion segmentation and Haar feature acceleration. Master of Engineering thesis, Dublin City University.

Full text available as:

[img]
Preview
PDF - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
6Mb

Abstract

A key challenge in a surveillance system is the object detection task. Object detection in general is a non-trivial problem. A sub-problem within the broader context of object detection which many researchers focus on is face detection. Numerous techniques have been proposed for face detection. One of the better performing algorithms is proposed by Viola et. al. This algorithm is based on Adaboost and uses Haar features to detect objects. The main reason for its popularity is very low false positive rates and the fact that the classifier network can be trained for any detection task. The use of Haar basis functions to represent key object features is the key to its success. The basis functions are organized as a network to form a strong classifier. To detect objects, this technique divides each input image into non-overlapping sub-windows and the strong classifier is applied to each sub-window to detect the presence of an object. The process is repeated at multiple scales of the input image to detect objects of various sizes. In this thesis we propose an object detection system that uses object segmentation as a preprocessing step. We use Mixture of Gaussians (MoG) proposed by Staffer et. al. for object segmentation. One key advantage with using segmentation to extract image regions of interest is that it reduces the number of search windows sent to detection task, thereby reducing the computational complexity and the execution time. Moreover, owing to the computational complexity of both the segmentation and detection algorithms we used in the system, we propose hardware architectures for accelerating key computationally intensive blocks. In this thesis we propose hardware architecture for MoG and also for a key compute intensive block within the adaboost algorithm corresponding to the Haar feature computation.

Item Type:Thesis (Master of Engineering)
Date of Award:November 2009
Refereed:No
Supervisor(s):O'Connor, Noel E.
Subjects:Engineering > Electronics
Computer Science > Multimedia systems
DCU Faculties and Centres:DCU Faculties and Schools > Faculty of Engineering and Computing > School of Electronic Engineering
Research Initiatives and Centres > Centre for Digital Video Processing (CDVP)
Research Initiatives and Centres > CLARITY: The Centre for Sensor Web Technologies
Use License:This item is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 3.0 License. View License
Funders:Enterprise Ireland, Science Foundation Ireland
ID Code:14856
Deposited On:12 Nov 2009 12:07 by Noel O'Connor. Last Modified 12 Nov 2009 12:07

Download statistics

Archive Staff Only: edit this record