Design Of Image Acquisition And Face Recognition System Based On MT7620A

With the great craze of artificial intelligence in recent years,the application area of image recognition is becoming wider than before.Face recognition,a biometric technique,has also become one of the hot topics.In the process of identity recognition,face recognition technology effectively combines personal information with the facial features of human beings.This feature makes it more safe,more natural,more fast than other identification technologies.However,the uncertainty of the realistic environmental factors and the diversity of the facial pose affect the calculation and analysis of the face recognition algorithm when recognizing the face,thus affecting the accuracy of face recognition.Therefore,in this thesis,we summarize the environmental applicability of different face recognition algorithms,that is,to find in which application scenario the algorithm can play its advantages and achieve the relative optimal recognizing the correctness of face.It is more important for the increasing human face recognition users.In this thesis,in order to analyze the recognition accuracy of a face recognition algorithm with different environmental factors,a face recognition system is designed as an experimental platform to analyze the algorithm.The system consists of video image acquisition and face recognition subsystems.The video image is mainly implemented with the hardware module,and the software module is mainly responsible for receiving and displaying the video stream data and realizing the face recognition function.The ultimate goal of our work is to test and analyze the experimental data based on the implementation of the face recognition algorithm using the system platform and to summarize the accuracy and environmental applicability of the face recognition algorithm when recognizing human faces with different environmental factors.Video image acquisition subsystem is built on an MT7620 A development board equipped with a USB camera to complete the video image acquisition,including a V4L2-based camera driver to capture the video image data,and a MJPG-streamer video stream server being responsible for processing the collected data and transfer it to the Qt Creator client.The face recognition subsystem performs face recognition in the video images.The real-time video surveillance platform based on Qt Creator client receives and displays the video stream data.In addition,OpenCV visual library is used to realize the face detection of video image data,sample training and target face recognition via the video surveillance platform.In this thesis,the goal of image acquisition and face recognition through cross-platform implementation is to establish a relatively complete "cloud platform",multiple video devices are only responsible for the collection of streaming data,data reception and processing unified by the "cloud platform" to complete This division of labor improves work efficiency while saving time and costs.In the thesis,three face recognition algorithms based on OpenCV,i.e.,Eigenfaces,Fisherfaces and LBPHFaceRecognizer,are chosen for comparative analysis.The face database in the experiment collects the faces of the surrounding students and friends and is used as the sample set,and the simulated environmental factors include light intensity,the angle of human face and occlusion of three,in order to achieve as much as possible to simulate realistic environment when the face recognition system used frequently in station security checks,intersection monitoring and company application scenarios.When analyzing the three face recognition algorithms for target face recognition,the experimental results show that the performance of Eigenfaces and Fisherfaces algorithms are basically the same under different environmental factors,and the accuracy of the target face recognition of Fisherfaces is slightly higher.LBPHFaceRecognizer is significantly better than the other two algorithms in some specific environments.Therefore,the user can combine the experimental conclusion to select a corresponding face recognition algorithm under the specific application scene for face recognition.