Real-time Face Detection Using Java Programming Language on Embedded Systems

Abstract

In the past half century growth of computer systems has made mankind’s life much easier than before. This growth in technology of computer systems improved our lifestyle in different criteria and developed our knowledge even faster than before. Quality management of products in mega factories to micro computers which are used in medical technologies to make surgery process faster and more accurate are just a few instances of this improvements. Embedded system is one of the usages of this new innovated technology for the past decades. An embedded system is a system that has everything which a computer needs to run, like processor and memory and IO (Input and Output), but in a very small scale. Though, the point of Embedded system is that it can run in unstable conditions; for example, in hot weather or under water or places that have enormous shakes or etc. On the other hand, one of the areas which has been very popular in computer field is photography and image processing. Different types of computers can be used, based on our requirements, to do jobs like noise reduction, pattern recognition, obtaining 3D pictures and in lots of other fascinating domains. Face detection, a kind of pattern detection, is a pre-requirement of face recognition that has enormous usage these days. This field has also been considered as one of the most popular research areas since annually more than a thousand of titles published in main engineering and scientific journals are allocated to this area, academic databases journals such as “Academic Search Premiers”, “IEEE Explore”, “SCOPUS”, and other important databases contain several research works in the field of face detection. The two main important issues in face detection processing are time consumption optimization with the intention of reaching highest speed possible while maintaining the second aspect which is an acceptable accuracy of the detection results. Thus, C and/or C++ are employed to do the detection process as fast as possible mainly because they are of good programming languages that are considered close to low level languages, in spite the fact that they are high level programming languages. Other high level programming languages have not been used in this area due to the fact that they have more latency for execution compared to these two languages. One of these popular high level programming languages is Java. Java is an object oriented programming language and it has lots of open source libraries that implements different classes for different purposes, which can be used by programmers. But the main thing that has made Java programming language the most popular one for software developers is that it is platform independent. For languages like C or C++ the source code should be compiled for each platform separately; consequently, the outputs of programs are not guaranteed to be the same for them as well. However, as programs written in Java run on a virtual machine and this virtual machine itself runs on top of operating systems, outputs of the programs are guaranteed to be the same for all of these operating systems. Actually, the only thing that should be provided is an interpreter which is the JVM (Java Virtual Machine) installed on top of the OS (Operating System). According to Barr & Frank [1], this virtual machine as an interpreter causes at least ten percent of latency to provide outputs for a program which is written in java compared to the same program that is written in C and/or C++. From embedded system point of view, it is also an issue as there is no software for JVM provided in this area. Fortunately, few processors have been developed recently which are able to execute java programs directly on the processor with no need of an interpreter. These processors implement JVM in hardware, which are mostly soft-cores, regards to research purpose, and sometimes hardcore architecture instead of a JVM application. As a result, it is expected that the mentioned extra latency which is caused by virtual machine interpreter is avoided for an embedded system that runs java programs through usage of hardware implemented JVMs. The aim of this research is to run a face detection application which is implemented by java on a hardware implementation of Java Virtual Machine (JVM) in an embedded system and compare it with the PC (Personal Computer) version of the same program. Moreover, problems that might occur in whole process will be inspected and analyzed. Finally, the result will be compared with a couple of other face detectors which was previously proposed by Sim and Yan as their Master of Engineering Thesis. First chapter gives a brief introduction towards the whole idea of research. A quick history of face detection systems is provided in this chapter and the outline of the thesis is explained as well. The second chapter brings an insight over current algorithms of face detection and it is followed by an explanation on the algorithm which has been used in this research in full details. Then in third chapter, we will talk about Java programming language and Java Virtual Machine and its implementations, especially hardware implementation. After all in this chapter, JOP (Java Optimized Processor) will be reviewed as a hardware implementation of JVM, the manner by which it has been used in this research. The forth chapter explains the face detection application which is implemented in Java by one of the open source libraries. In the fifth chapter the very same program will be mapped to the JOP and implemented by DE2-115 FPGA board. Eventually, I will bring future works and possible improvement for the research in the last chapter.