Face Recognition And Application Research Based On Deep Learning

Face recognition technology is one of the popular computer vision applications.Deep learning,particularly the convolution of the neural network,has resulted in significant progress in facial recognition in the recognition of static images.Nevertheless,there are many problems still facing complex image recognition and video stream recognition.The detected face is often influenced by several factors,such as angle of shot,location,lighting,low resolution,etc.,which render the following face recognition very difficult.In security surveillance,investigation,smart cities,and other fields,dynamic face recognition technology is commonly used,and therefore has high science value.In order to address the above issues,this paper uses the deep learning approach to complete work on face recognition and integrates face recognition with different application scenarios.The research material of this paper is:Designed and implemented a dynamic face detection algorithm based on MTCNN network.The MTCNN face detection algorithm is better than VJ face detection algorithm in detection angle,posture,multi-face detection,etc.,relative to standard VJ facial detection algorithms.MTCNN face detection algorithm is a three-layer cascade network with an untracking structure and a low performance in real-time only in the CPU operating environment of ordinary PCs.To solve the above problems,the enhanced Camshift algorithm is used for face-tracking and the MTCNN facial system is used to configure the Camshift tracking algorithm's search window location and scale.In addition to the original Camshift algorithm,facial morphological constraints and Kalman filters solve near color interference problems and transfer target tracking.The experimental findings demonstrate that the frame rate is about 56 FPS after combining the enhanced tracking algorithm for Camshift with the MTCNN face detection algorithm to fulfill the real time requirement.The Inception-Res Net v1 face recognition network and the Mobile Net v3 mask recognition network are developed and implemented.The face picture obtained by MTCNN facial detection algorithm is then scaled to match the detected facial picture with binocular rotation and affinity transformation,providing the basis for a subsequent network InceptionRes Net v1 to extract facial features.After scaling,aligning and comparing the images captured by the camera with the database,the recognition results are achieved and the face recognition process is completed.Mask identification is one of the facial recognition technology situations.The Mobile Net v3 network is used for the identification and recognition of masks.Face detection and synchronization between the mask face and the mask-free face image is performed.This is the basis for developing the mask recognition data set and using the Mobile Net v3 network for model training and recognition.The experimental results show that mask accuracy is more than 98 percent.In combination with Keras,the Tensor Flow deep learning framework and the Open CV Vision Library,the Pycharm developer environment and the PyQt5 design software interface for a face detection and recognition system,the entire system consists of three modules: face recognition,face recognition,and mask recognition.Experimental tests show that the entire face detection and recognition system can complete face detection,facial recognition and mask recognition functions.