Background subtraction using coplanar filter and Quadtree decomposition for objects counting

Other Title(s)

طرح الخلفية باستخدام مصفاة كوبلنر و شجرة التحليل الرباعي لعد الكائنات

Dissertant

al-Zaghal, Fayiz Kamal

Thesis advisor

al-Hammuz, Sadiq

Comitee Members

Kayid, Ahmad
Uwais, Suhayl

University

Middle East University

Faculty

Faculty of Information Technology

Department

Department of Computer Information Systems

University Country

Jordan

Degree

Master

Degree Date

2015

English Abstract

Traditional background subtraction algorithms are used mainly to discover objects in images by subtracting them from known background images for same scenes excluding these objects.

However, these traditional algorithms fail in detecting all edge pixels, which is in turn influences the accuracy of resulted detected objects.

Therefore, this thesis introduces an enhancement for the traditional background subtraction algorithms, considering applying two techniques within segmentation tool; coplanar filter; to improve the detection of all edge pixels, and Quadtree Decomposition; to divide images into homogenous blocks.

Both algorithms; the enhanced, and traditional are then applied to design a car tracking system using MATLAB to detect and count the number of cars in a specific street.

The number of detected cars resulted using each algorithm is compared later with the actual number of cars in that street for performance evaluation purposes.

The evaluation is conducted based on detecting and counting the number of cars both in; video frames within the online stage or uploaded images from a dataset within the offline stage.

After that, comparing these frames (images) with a background image for a street, which is devoided of cars.

The threshold of the proposed system will be adaptive over each segment of the image, normally threshold for traditional background subtraction is 0.5 where any pixel value greater than 0.5 assumed to be white while lower than 0.5 is assumed to be black.

This research results illustrate that the enhanced background subtraction algorithm outperforms the traditional algorithm in counting the number of cars in all frames.

This system achieved 47.01% average X accuracy rate for the traditional background subtraction algorithm and 81.19% average accuracy rate for the enhanced algorithm.

Main Subjects

Information Technology and Computer Science

No. of Pages

64

Table of Contents

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-721231