Design Of Face Recognition Embedded System Based On DSP

As the development of society, people puts forward higher requirements to the accuracy, security and real-time of personal identification ,traditional methods no longer meet the requirements, while biometric identification technology obtains fast development and wide application with its own characteristics and advantages. Compared to other biometric identification such as fingerprint, voice, palm-print and so on, face recognition technology provides a way of identification which is direct, friendly, convenient, non-invasion, high reliability and stability,these advantages make face recognition have great potentialities in many fields such as personal identification, automatic monitor and control and human-computer interaction and so on.Based on requirement analysis and experiment, this paper firstly presented the whole realization scheme of system, the design of the whole system included hardware configuration of DSP, the study of related algorithms of face recognition and its realization on DSP, then used SEED-VPM642 DSP development board as the hardware development platform, transplanted the algorithms of face detection and face recognition into DSP, and then constructed the face recognition system based on DSP. The main contents of this paper are as follow:1. Deeply studied the algorithm principle of Adaboost, respectively realized training process of classifier used for face detection and detecting process on-line, and transplanted it into the face recognition system of DSP.2. Improved the algorithm of face recognition, put forward a method combined with bidirectional two-dimensional principal component analysis and PCA, that is, firstly extracting features in the direction of row and column in turn directly on two-dimensional image, then on the basis of which compressing feature dimensions using principal component analysis, this method not only guarantes high recognition rate but also saves storage space for face features.3. In the integrated development environment CCS of DSP, the work of system is divided into five tasks which are image acquisition, image processing, image display, network initialization and network communication between PC and DSP, then the author configured these five tasks using real-time operating system DSP/BIOS in CCS, realized proper scheduling among these tasks.4. Realized real-time acquisition of face image using CCD camera controlled by DSP, established system's own face feature database, and stored them on PC, realized communication between DSP and PC and real-time configuration by network interface of DM642.In the prophase algorithm verification, it showed that the algorithm of face detection based on Adaboost met the requirement of real-time and efficiency on the basis of front face database of CMU/MIT; Testing on the face database of ORL using improved method of face recognition, it showed that the highest value of face recognition rate reached up 97.5%, through many times and repeated parameter adjustment; After completed construction of face recognition on DSP, the author tested in real-time on this system, the result further verified the rationality of system's construction, achieved expected objective in the aspect of real-time, accuracy and security.