Face Recognition On Embedded System With Dual Core Architecture

As a convenient and quick biometric technology,face recognition is increasingly used in the field of identity authentication.In recent years,face recognition systems based on deep learning have developed rapidly.However,face recognition algorithms based on deep learning mostly run on server platforms.However,providing cloud computing interfaces for embedded devices has high development costs.At the same time,it is restricted by the network and cannot meet people's requirements for ease of use.Therefore,the research on the embedded face recognition system that is easy to use and can be trained and identified offline has important significance.The current embedded face recognition system still has a situation where the detection speed of the key points of the face is slow and the recognition rate is low under the single-sample training scenario.In this paper,the design and implementation of face recognition system are deeply studied under the embedded platform of dual-core architecture.First of all,because the face key point detection classifier provided by Opencv runs slowly on the embedded platform,this paper uses the Adaboost algorithm to train key classifiers dedicated to the embedded platform.On the basis of face detection,combined with the coarse positioning of key points,fast and accurate localization of key points on the face is achieved.Second,the traditional face recognition algorithm has poor recognition performance in single-sample training.This paper proposes an improved joint Bayesian face recognition algorithm for single-sample face recognition.This algorithm combines the method of adaptive enhancement and feature fusion of face data.When the training sample has only one image,the adaptive expansion of the face data is performed.After the global features and local features were extracted from the sample,the joint Bayesian algorithm was used to train the model.The method of classification using global and local classifiers is used for classification.In the case of one-sample training data,this method can accurately recognize face changes in face.Finally,in terms of system software implementation,the company's XDC algorithm tools and software system architecture are fully utilized to design the software,and the dual-core operating resources are reasonably allocated to optimize the porting and operation of the algorithm.Using system-integrated infrared receiver modules and OSD controllers and other peripheral devices,the available human-computer interaction functions are realized.Finally,on the DM6446 platform,a face recognition system based on an improved joint Bayesian algorithm was implemented.The experimental verification shows that the face recognition system of this paper has good man-machine interaction function and can accurately and quickly identify faces.