Video Face Recognition System Based On Depth Network Design And Implementation

Face recognition has always been an important research hotspot and difficulty in the field of computer vision.It is a biometric recognition technology based on human face feature information.In recent years,with the rapid development of related technologies in the field of computer vision,especially the continuous improvement of the theory of deep learning technology and its wide application in various fields,the use of deep learning to solve face recognition problems has attracted more and more attention.At present,most of the face recognition systems on the market are static face recognition systems.Static face recognition has the advantages of natural intuition,non-contact and concurrency,but it requires the tester to stand close to the camera sensor and wait for the recognition results,which brings many inconveniences to the tester.Compared with the traditional non-contact static face recognition system,we designed a video face recognition system.The system has the advantages of convenient identification,simultaneous multi-person identification,fast identification and accurate identification.There are three main difficulties in video face recognition system: face difference(facial expression,illumination,posture,occlusion,motion blurring in unrestricted environment);accuracy;real-time.To solve the above problems,the system we designed includes four modules: face quality assessment,face detection,face feature extraction and face matching.Face quality assessment includes pose detection and face blur detection.Through pose detection and face blur detection,low-quality video frames with larger pose and blurred face are filtered,while high-quality video frames with smaller pose and clearer face are retained.Face detection is a real-time video frame for face detection.In this system,the depth network MTCNN model is used for face detection.Face feature extraction is to extract tailored face features.In this system,inception?resnet?v1 model is used to extract face features.In order to extract discriminant features,we combine embedding layer and flatten layer features to retain more face features and learn robust face representation.Face matching algorithm refers to the similarity calculation between the input image and the face image in the matching database to realize face matching.In order to achieve precise face matching,the system uses two video streams for joint face recognition,which is more accurate than single frame face recognition.In order to realize real-time face recognition,the system adopts the strategy of skipping frames,which can reduce the operation cost of the system and accelerate the recognition speed.We construct an adult face recognition system based on four modules: face quality assessment,face detection,face feature extraction and face matching,and verify that the system can realize real-time face recognition in an unrestricted environment through system functional testing.Our main work is as follows.A video-based face recognition system is designed and implemented.The system consists of four modules: face quality assessment,face detection,face feature extraction and face matching.The innovations of our work are as follows: 1)Compared with the traditional face recognition system,the designed face recognition system is based on deep learning network,which can learn discriminant features and improve the accuracy of face recognition.2)The video face recognition system designed by us can realize face recognition under the condition of face moving.Compared with the traditional non-contact static face recognition system,the system has the advantages of convenient recognition,simultaneous multi-person recognition,fast recognition and accurate recognition.3)In the feature extraction stage,we integrate the multi-layer features of deep network to form a more robust face representation,which improves the accuracy of face recognition.4)In the phase of face recognition,we use multi-frame video stream data to make joint face recognition judgment,which improves the recognition accuracy of the system.5)In order to realize real-time face recognition,the system adopts frame-hopping strategy.The frame-hopping processing can reduce the redundant information of face in continuous frames,thus reducing the system operation overhead and speeding up the recognition speed.