Date of Publication
12-2018
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Electronics and Communications Engineering
Subject Categories
Electrical and Computer Engineering
College
Gokongwei College of Engineering
Department/Unit
Electronics and Communications Engineering
Thesis Adviser
Argel A. Bandala
Defense Panel Chair
Ryan Rhay P. Vicerra
Defense Panel Member
Elmer P. Dadios
Laurence A. Gan Lim
Jennifer C. De La Cruz
Raouf Naguib
Abstract/Summary
Computer vision (CV) is a scientific discipline that assists in the building of systems that obtains relevant information from images or multi-dimensional data. One of its major research area of concern is target detection and recognition, which involves feature extraction, classification and localization. For the past few decades different CV researches were focus on target (object) detection. However, as literature suggests, vision – based target detection and recognition introduces challenging problems due two important aspects: first, problems arising from complexity of images which are very rich in variances and diversities; such as rotation and scale brought about by the multi-view perspective that a certain vision sensor accumulates; and second, problems in terms of accuracy and speed for target classification. While traditional method of target detection and classification have proven to be effective there is a need to explore other methods and novel ideas to continuously extends it to a real-time application that is robust to any dynamic changes. Thus, the motivation of this research was to develop a recognition system which, together with a modified classification algorithm, generates sufficient and significant keypoints of targets using dynamic agent in the presence of any change in scale and rotation. To achieve this objective, two methods were introduced in this research: a novel method of Lagrange Learning Algorithm (LLA) in predicting other key features for detection of a target that is subjected to scale and rotation variability; and a modified Support Vector Machine which uses compound kernel function for classification for both linear and non-linear values. The system output accuracy was measured at 93.33%.
Abstract Format
html
Language
English
Format
Electronic
Accession Number
CDTG007968
Keywords
Computer vision; Image converters
Recommended Citation
Roxas, E. A. (2018). Vision-based target detection using modified scale and rotation invariant technique for dynamic agent. Retrieved from https://animorepository.dlsu.edu.ph/etd_doctoral/1449
Upload Full Text
wf_yes
Embargo Period
1-10-2023