| We have been in information epoch nowadays; our lives become digital more and more, everyone of us is also become digital, because the biological recognition such as fingerprint recognition, voice recognition, face recognition and retina recognition are very mature now. Biological recognition use the body characteristic which can't be duplicated to complete a verification of a body, it first abstract the characteristics of human body from a person, then convert them into the numerical information and store these information into computer, last use matching algorithm to verify the person.The traditional methods of identification such as passwords, certificates, IC cards etc, which can be counterfeited, peculated and decrypted because they can be separated from the owner is not safe. And the body characteristics is connatural and exclusive to a person, they are can be duplicated, stolen and leave. So the traditional methods of identification have been replaced by because the biological recognition steps by step.Embedded system is a special computer, which focus on application a and based on computer technique and its software and hardware can be cut. Its aim is to satisfy the consumption, function, security, cost and physical volume required by special application system.DSP is a processor which has special architecture. It can execute a DSP algorithm rapidly because of its Harvard structure which separates data from program, special hardware multiplier, pipelining and special DSP instruction. The traditional face recognition system are achieved depending on computers, both drivers of peripherals and image capturing depend on the hardware and the operating system of a computer, these result in some disadvantages such as the high cost, poor function in real time, and can't be fit to common consumer etc. The system which is referred in this paper is based on embedded system technology and use DSP in the face recognition has some strongpoint such as high rapidity, low cost and high security etc. In this paper we will introduce how to design a face recognition system based on ADSP-BF532, such as how to design drivers, how to porting code and protocol, how to optimizethe system etc.The face recognition system referred in this paper can be used in gating and checking on work attendance. The system has a camera which can capture color VGA image and a LCD which displays the image captured by the camera real-time. User can adjust his face position in the image displayed by LCD to be easy to recognize his face and use a keyboard to control the recognition progress. The result of recognition will be displayed on LCD, and transmits to remote Personal Computer by Internet.The system design is parted into three parts; they are hardware design, driver design and software design. In hardware design, we part the system into six modules at hardware layer; they are kernel module, logic module, capture module, display module, network module and memory module. In every module, we will introduce its function and chip-selection. In driver design, we will introduce how to initialize the peripherals, and how to code correct read-write programs which used by ADSP-BF532 to communicate with the peripherals; we also introduce the inside logic of CPLD which used to extend the peripheral interface of ADSP-BF532. In software design, we will introduce the work-flow of the system, recognition algorithm selection and protocol selection.Code porting is composed of algorithm porting and protocol porting. The recognition algorithm is implemented by C++ language in Visual C++ develop environment, Visual DSP++ also supports C++ language, but compared to C language, C++ is lower efficiency than C. So first the algorithm must be converted from C++ language to C language, then ported into ADSP-BF532. The algorithm porting parts two steps, the first step is C++-to-C conversion, the second step is configuration of memory, especially configuration of heap.In protocol porting, we only port the UDP protocol of LwIP based on the system's requirements. In this part, we will introduce the implementation of every layer of UDP protocol, especially the driver design of CS8900A which is belonged to network-interface layer of UDP protocol.Optimization is composed of software hardware and optimization optimization. The system after ported has some problems, such as slow image capturing, image data losing and LCD can't display real-time, the running speed of the system also doesn't satisfy the requirements, so system optimization is verynecessary. In the hardware optimization, we first adjust the read-write timing of asynchronous memory bank which the address of camera is mapped into, and then we use cache on the image memory which is read and write frequently. The software optimization is composed of the program structure optimization, turning on Visual DSP++ optimization option, RAW RGB-to-RGB24 converting function optimization and simplifying the interrupt program of camera.We do lots of experiments after all work done, the results turn up trumps. The image captured by camera is very clear, its color is also correct. The running speed of the system after optimization is improved 19.64% compared to the system which isn't optimized. LCD display rate is 6.25 f/s, it is already a good result of the processor which has single-kernel and parallel interface. The real-time performance is also satisfied the requirements. After UDP protocol porting completed, CS8900A can communicate with remote PC normally, the image date is not lost during transmission. |
PDF Full Text Request