Research Of Face Recognition Based On Embedded Intelligent Monitoring System

Automatic face recognition (AFR) technology is a challenging task. It involves image processing, pattern recognition, computer vision, neural network and other subjects. The research of the embedded face recognition based on intelligent monitoring system is built on the embedded operating system and embedded hardware platform with characteristics of high starting point, new concept and practical.In monitoring, face recognition will be affected by light and gesture. The recognition accuracy and recognition speed is still difficult to meet people's expectations. This paper is looking into the robust facial feature description and efficient face recognition algorithm on embedded intelligent monitoring system.Firstly, introduce the embedded Linux operating system and GM8180 platform architecture. Explore key issues in the design of embedded face recognition system. As a result of embedded resource-constrained, it is not only need to consider the high recognition rate, but also the algorithm complexity. We select EPL as the embedded face recognition algorithm which has a lower complexity and linear dimensionality reduction.Secondly, as the parametric Gabor wavelet has a good match with the simple cell receptive field model, we use Gabor wavelet to represent faces. Consider transformed vectors as independent samples can increase the number of samples. We also remove the duplication of the Gabor wavelet to ensure the recognition rate.Finally, build AFR system based on the foregoing algorithm with the histogram equalization and the classifier of cosine angle between each subspace. Then in this system, fast PCA algorithm and floating-point to fixed-point method is used to optimize the speed of face recognition.In this paper, we use local Gabor wavelet to represent faces. The experimental result shows that the EPL algorithm based on local Gabor wavelet is better than the directly EPL algorithm. In small samples and small dimensions, the recognition rate can reach 95%. What's more, fast PCA algorithm and floating-point to fixed-point method can improve the recognition speed. The speedup of training process and identification process is 4.93 and 2.55, respectively. The propose face recognition method not only has a theoretical value, but also has a reference significance for the application development of embedded system and AFR system.