Research And Implementation Based On Embedded Face Recognition Algorithm

At present,the Advanced learning face recognition system is very high in hardware,and it is not suitable for the embedded devices with low matching.The application of embedded technology and the improvement of the algorithm are advantageous to the efficient and flexible use of embedded face recognition system.The main work of this paper is as follows:1.In this paper,we discuss the methods of face detection and face recognition in embedded devices.In face detection,two main classifiers,the principle of haar classifier and LBP classifier,and the final selection of cascaded strong classifier are analyzed.In the face recognition,in addition to the analysis of PCA algorithm,LDA algorithm and LBP algorithm,the modified LBP operator is used to complete the face recognition system by combining the rotation invariant LBP operator and the block weighted LBP operator.2.The selection of the hardware of face recognition system and the construction of software environment are introduced emphatically,these hardware and software environment is the basis of the design of face recognition system,and it is the important guarantee to complete the project.The hardware chooses the ARM series CORTEX-A9 Architecture Development Board,the Software Development environment construction,completes the cross compiles,the OpenCV transplant and the QT transplant and so on.3.The principle and method of face detection and recognition in embedded platform and the design of softxware and hardware are emphatically analyzed.The design of embedded software system is mainly divided into four modules.They are based on QT graphical user interface development,based on V4L video acquisition and implementation,the use of haar+adaboost cascade classifier for face detection and the use of local fusion LBP algorithm for face recognition,and contrast eigenfaces and fisherfaces algorithm.4.The face recognition system based on embedded is completed independently,including the establishment of face database,the training of feature library,the human face testing and recognition.Through some test data analysis,this system uses the algorithm to be able to carry on the very good face recognition in the ORL,the Yale as well as the self built human face storehouse.Through the experimental data to know:(a)Haar classifier face detection is slower than LBP classifier to detect human face,but its recognition rate is high;(b)Three face recognition algorithms have their advantages and disadvantages,when training recognition,LBP recognition algorithm is relatively slow,arm platform,The rate of LBP recognition is higher than that of PCA and LDA;(c)When the system is being used for face recognition,for the resolution of the image,the higher the resolution,the better the recognition rate;(d)for different human face database,the human face database is relatively small,its recognition speed is faster,but the identification test sample will be smaller,the recognition rate will be correspondingly reduced;(e)for the improved LBP operator,using the algorithm of block weighting to face recognition,the higher the recognition rate of the block,but the longer the time of recognition,the comprehensive time and the recognition rate,the final selection of this paper is divided into four pieces for face recognition.Through the experiment conclusion,the system has a good robustness for illumination,can also be well recognized for wearing glasses and so on,the whole time delay 28.50ms in the video streaming system has a considerable stability and reliability,the design of the face recognition system has certain practical value.