Design And Implementation Of Face Recognition System Based On Monitoring Scenarios

With the rapid development of modern science and technology,the concept of artificial intelligence is becoming more and more popular.In our daily life,the application scenes of artificial intelligence technology are increasing and one of the most representative of the technology is face recognition technology.Face recognition technology research is not restricted within laboratory only,but to continue to make important progress and achieve its greater value in people's daily production.For example,specific occasions use face recognition access control system,public security or customs use face recognition technology to authenticate specific objects,computer or mobile phone devices scan face to unlock,payment software identify face to process payment function,etc.However,these face recognition techniques are used in constraint application scenes,which means the objects are cooperative and front face images are obtained under a certain controlled scene.In the era of information and data explosion,video surveillance systems have penetrated into every corner of our lives.Out of the increasingly strong needs of security,more and more industries such as city construction and transportation need intelligent monitoring system to support all kinds of job,so that the researches of intelligent monitoring system with face recognition technology have drawn more and more attention.Under the scenarios of surveillance,face recognition technology is facing the challenges of non-constraint conditions.In monitoring videos,objects would not initiatively face the camera therefore resulting in various face poses and gestures.Even block and occlusion problems would occur because of wearing glasses,hats and other ornaments.In addition,changes in lighting under the monitoring scene will also affect the recognition.Taking into account that the data quality of the monitoring video is limited by the hardware condition,the low resolution of face image data in the video would further increase the difficulty of the face recognition task.Therefore,it is a very meaningful topic to design and implementation a face recognition system based on surveillance scenarios.This work is not only an important research in the field of security,but also a valued research that can be applied to a wider range of people's livelihoods.Based on previous researches,this paper proposes a face recognition algorithm based on multi-scale completed local binary pattern and a Caffe-based deep learning network to fulfil the task of face recognition.And numerous experiments are conducted to verify the effectiveness of the proposed algorithm and network on FERET,CASIA and LFW datasets.On this basis,an integrated face recognition interactive system is implemented to demonstrate the work.Specifically,the research contents of this paper include the following three parts:1.Research on traditional face recognition feature descriptorsThe design and implementation of the traditional face feature descriptor requires a strong prior knowledge,so that the descriptor can effectively represent the feature of original image.This paper proposes a feature descriptor based on multi-scale completed local binary model.Features are input to the classifier for recognition and matching afterwards.Experiments are carried out to verify the effectiveness of the proposed recognition algorithm on ORL Faces,Yale Faces,FERET dataset.2.The construction and training of deep learning networkBased on the study of the deep convolution neural network,an improved deep res-network used for face recognition is designed and implemented.By using public CASIA WebFace and LFW databases to train the network and making comparison with other state-of-art network model to verify the feasibility and effectiveness of the proposed network.3.Implementation of face recognition systemThe face recognition system based on the monitoring scenario includes the following main modules:data acquisition,face detection and face recognition.The system can both receive real-time video data and local video data.And a user-friendly GUI is implemented to show the achievement of this paper work.In summary,this paper mainly studies the face recognition algorithm under the monitoring scene and implements the interactive recognition system.Two face recognition algorithms are proposed and various experiments are conducted to verify the feasibility and effectiveness of the algorithms.At last,a usable system is implemented.