High Performance Low Power Embedded Vision Systems
Technical University of Crete, Electronic and Computer Engineering Department, Microprocessor & Hardware Laboratory
Technical University of Crete, 2013
@phdthesis{nikitakis2013high,
title={High Performance Low Power Embedded Vision Systems},
author={Nikitakis, Antonis S.},
year={2013}
}
Human vision is a complex combination of physical, psychological and neurological processes that allow us to interact with our environment. We use vision effortlessly to detect, identify and track objects, to navigate and to create a conceptual map of our surroundings. The goal of computer vision is to design computer systems that are capable of performing these tasks both accurately and efficiently at real-time and using limited resources. There are numerous computer vision systems today in various application fields that demand very fast and accurate detection of certain points in video or live streams. Considering the multimedia and the upcoming wearable computing field, one very important challenge is the implementation of fast and detailed Object Detection/Tracking and object Recognition systems. In particular, in those systems, it is highly desirable to detect and locate certain objects within a video frame in real time but also using minimum energy. Although a significant number of Object Detection, Tracking and Recognition schemes have been developed and implemented, triggering very accurate results, the vast majority of them cannot be applied in state-of-the-art mobile multimedia devices; this is mainly due to the fact that they are highly complex schemes that require a significant amount of processing power, while they are also time consuming and very power hungry. In this thesis we present three different approaches in building high performance Embedded Vision Systems, while our focus is both real time performance as well as low power consumption. Following a certain set of principles identified after the analysis of numerous vision system we have implemented three state of the art computer vision schemes utilizing state-of-the-art FPGA devices along with embedded processors: The OpenSurf which is a feature extraction algorithm, the RFCH which is a state of the art object detection algorithm, and the OpenTLD which is a self-trained stable tracking algorithm.
July 22, 2014 by hgpu
