Embedded operating system optimization for face recognition system

Abstract

Compiler optimizations have been proven to be beneficial in improving performance and efficiency of binary codes across different type of operating systems. Improvement from optimization of program code during compilation are obvious to the system performance, memory performance, disk access, energy efficiency and nearly all aspect of the system. This is particularly attractive to an embedded system where performance and efficiency are seriously considered. This research focuses on implementing those proven compiler optimizations to the development of an embedded operating system and studying the effect of such implementation to the performance and efficiency of the system. This thesis outlines the methods available to develop a custom operating based on the Linux kernel and ways to implement such optimization during the compilation and development process. Focusing on face recognition systems implemented on embedded Single Board Computers, optimization effects to an embedded operating system are studied through software and hardware perspective. From the software perspective; booting performance, code execution performance and binary efficiency of different prototypes developed with varying level of compiler optimization are tested and examined. From hardware side; system memory footprint, current draw, power consumption and energy usage of the different prototypes are tested and measured. Results obtained in this thesis shows that compiler optimization are beneficial not only when applied to the program code, but also have significant effects when applied to the operating system.