Please use this identifier to cite or link to this item: https://ptsldigital.ukm.my/jspui/handle/123456789/457806
Title: Moving object detection for smart video surveillance system
Authors: Mohammad Hedayati (P53341)
Supervisor: Wan Mimi Diyana Wan Zaki
Keywords: Signal processing -- Digital techniques
Computer vision
Video recording
Motion detectors
Electronic surveillance
Universiti Kebangsaan Malaysia -- Dissertations
Dissertations, Academic -- Malaysia
Issue Date: 28-Aug-2012
Description: Moving object segmentation is a very important step in many computer vision applications like video game, human- computer interface and surveillance security system. The background subtraction (BGS) technique is the most common approach for real time object segmentation. Typically, most BGS techniques work reasonably well in simple environments. This is because each pixel is treated individually without considering its neighbouring area. However, their performances are highly sensitive with respect to environmental variations like illumination change and small background fluctuation. The aim of this research is twofold that is, first to review the most frequently used BGS method and perform analysis based on processing speed (PS), memory usage (MU) and segmentation accuracy (SA). Second, is to propose and develop a novel approach for object segmentation to overcome the drawback of the common BGS algorithm. Five popular algorithms are investigated which include median-filtering (MF), approximate median (AM), running Gaussian average (RGA), Gaussian mixture model (GMM) and Kernel density estimation (KDE). The aforementioned algorithms are subjected to four challenging environments in order to test its performance. Implementation is done by using the MATLAB 2009a software and Open Source Computer Vision (OPENCV) library on the 3.2 GHz CPU with 5 GB RAM computer. Next, a new BGS and object detection method for a real-time video application using a combination of frame differencing and a scale-invariant feature detector is proposed. The new method combines the benefits of background modeling and the invariant feature detector and as such, the method is expected to improve the segmentation accuracy. The preliminary investigation reveals that the Gaussian algorithms namely GMM and RGA yield the two most balance results in three out of four scenario with segmentation accuracy of 82% for GMM and 87% for RGA, processing speed of 1.6 fps for GMM and 20 fps for RGA, and memory usage of 2304KB for GMM and 1094KB for RGA. Subsequently, the performance of the proposed method is compared with that of the GMM which showed that correct classification can be increased up to 98.7% with improvement in both segmentation and computational time since the speed of keypoint model is dependent on the number of features recognized in the scene and not based on individual pixels. Besides, it is noted that the proposed method has shown great potential in handling variation of the environmental challenges like shadow effect and lighting fluctuation. Thus, in short, the developed method is able to overcome the drawback of the traditional BGS method.,Master of Science,Certification of Master's / Doctoral Thesis" is not available"
Pages: 91
Call Number: TK7882.M68M844 2012 3 tesis
Publisher: UKM, Bangi
Appears in Collections:Faculty of Engineering and Built Environment / Fakulti Kejuruteraan dan Alam Bina

Files in This Item:
File Description SizeFormat 
ukmvital_122297+SOURCE1+SOURCE1.0.PDF
  Restricted Access
11.49 MBAdobe PDFThumbnail
View/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.