The Department of Electrical and Computer Engineering invites you to the PhD Thesis Defense of Christos Kyrkou for the degree of PhD in Computer Engineering
Wednesday 15th January 2014, 10:00 a.m.
Room ΧΩΔ 02 – 016, New Campus, University of Cyprus
Abstract:
Intelligent embedded vision systems rely on learning algorithms to process visual information and are becoming increasingly important for emerging applications in security and surveillance,
automotive, and robotics amongst others. An integral part of vision systems is the ability to visually detect the presence of objects. This enables them to understand and interpret information from their environment and is critical towards building systems that exhibit increased contextual awareness and are able to adapt and react more intelligently to their host
environment. Object detection involves the extraction of information from an image, and processing of the information by a machine learning algorithm to determine whether that
information corresponds to an object of interest. This overall process is computationally intensive primarily for two reasons: a) The number of search windows that are generated from
a single image and need to be classified can grow exponentially with the image size. b) The computational complexity of the underlying machine learning algorithm used for classification.
This makes it challenging for software implementations to achieve real-time performance (over frames-per second) without sacrificing detection accuracy; hence urgent needs for acceleration arise. Also, considering that embedded object detection systems need to operate under low power constraints and in most cases with limited hardware budget there is a need for dedicated hardware architectures that are optimized for the specific embedded application.
Consequently, the contributions of this thesis relate to the development of real-time hardware architectures for Support Vector Machines (SVMs) and the Viola-Jones cascade classifier, two
widely used algorithms for visual object detection, in an attempt to tackle the above issues. In addition the work in this thesis attempts to improve the performance of object detection
systems by reducing the number of windows that are generated from an image. Hence, a hardware architecture is proposed that integrates together edge image processing, processing and classification in order to provide real-time object detection. The proposed architectures are experimentally evaluated on Field Programmable Gate Arrays (FPGAs) computing platforms demonstrating real-time performance (over frames-per second) that enables them to be used in real-time object detection applications.
Bio:
Christos Kyrkou received his Bachelor’s Degree and Master’s Degree in Computer Engineering from the University of Cyprus in 2008 and 2010, respectively. Since 2010 he is a Ph.D.
candidate in Computer Engineering at the University of Cyprus. He is a researcher at the Embedded and Application Specific System-on-Chip Laboratory (EASoC) at the KIOS research
center. His research interests include embedded systems, digital hardware architecture, computer arithmetic, Field Programmable Gate Arrays (FPGAs), computer vision, and pattern
recognition. He is involved in various projects funded by the European Commission and the Research Promotion Foundation of Cyprus. He is a student Member of IEEE and ACM and also
a member of the Technical Chamber of Cyprus (ETEK).
