Citation

Ibrahim, Amalina (2018) Silhouette extraction from gait videos with illumination variates. Masters thesis, Multimedia University.

Full text not available from this repository.

Abstract

In recent years, various methods have been invented for video data acquisition in biometric. The final outcome of it is for authentication, where it aims to solve identity mistaken issue of person recognition. The main target in video data acquisition is to achieve higher rate of data accuracy and correctness. Computational time is similarly among one of the important targets as it represents processing efficiency and complexity. Unfortunately based on high relative fact, accuracy and time are not cohabiting well. However, background subtraction is generally based on a static background hypothesis which is often not applicable in dynamic illumination environments. With indoor scenes, reflections or animated images on screens lead to background changes. In the same way, due to wind, rain or illumination changes brought by weather, static backgrounds methods have difficulties with outdoor scenes. Thus, dynamic illumination can bring such a serious either positive or negative impact towards biometric person recognition, where accuracy may be compromised. In other words, varying illumination causes challenges for gait silhouette extraction from videos used in person recognition problem. Therefore, the following objectives and deliverables are set in this work is to develop a pixel-based adjustment method to extract gait silhouette from videos with illumination variates on micro and macro based decisions. After that is to evaluate the performance of the proposed gait silhouette extraction method. There are several known techniques based on background subtraction and one of them is Mixture of Gaussian (MOG) that is used in this research. The MOG has been widely used for robustly modelling complicated backgrounds, especially those with trivial repetitive movements. Thus, this research combines several methods and created three modules; a prior processing module, an illumination modelling module, and motion extraction module. The prior processing module monitors robustness of the data to accommodate the next following modules. For the illumination modelling module, it manipulates pixel values in each frame to increase accuracy of extracting potential features in the next motion extraction module. The implementation of these two proposed modules will produce a silhouette extraction framework which adaptive to uncontrolled light scenes. The extracted features will then be captured and used for persons recognition purposes. Through the proposed methods, this research has found that gait silhouettes from videos containing illumination inconsistency were able to be extracted and used in person recognition task. The proposed pre-processing image enhancements capable of removing minor noises and capture the moving object in detail during silhouette extraction regardless the condition of brightness level. By having good silhouette quality, it leads to possibility of having good recognition rate during gait recognition phase. Indirectly, the performance of extracted silhouette is validated by the correct classification rate during recognition process.

Item Type:	Thesis (Masters)
Additional Information:	Call No.: TK7882.P7 A43 2018
Uncontrolled Keywords:	Human activity recognition
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7800-8360 Electronics
Divisions:	Faculty of Computing and Informatics (FCI)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	18 Apr 2023 07:32
Last Modified:	18 Apr 2023 07:32
URII:	http://shdl.mmu.edu.my/id/eprint/11358

Downloads

Edit (login required)