Design And Implementation Of Real-time Face Recognition System Based On OpenCV

With the development of Internet of things and artificial intelligence,face recognition system based on embedded system has been widely used.Although the face recognition system can get very high recognition results under controllable conditions,however,in complex scenes such as streets and shopping malls,the collected face images are easily affected by the factors such as attitude,expression,occlusion and illumination.The features that can be extracted are very few or even difficult to extract,which brings certain resistance to face recognition.This paper studies the design and implementation of face recognition system in complex environment by using the method of fusing global feature and local feature,and realizes the purpose of rapid and accurate recognition of face identity under unrestricted conditions.Face recognition mainly includes three key links: face location detection,feature extraction and feature comparison.In this paper,for face location detection in complex environment,MTCNN face detection algorithm is selected,which can not only locate face quickly,but also detect face under occlusion and illumination conditions.For feature extraction,in order to solve the key problem that illumination and occlusion environment can not extract global features accurately,a comprehensive analysis method with global features as the main and local features as the auxiliary is adopted.The optimized method of fuse global feature extraction based on Squeeze Net network and LBP and gradient local feature extraction based on image block method is proposed in this paper,which can improve the accuracy of facial feature extraction.For the comparison of recognition,Euclidean distance is used to calculate the similarity between the face in the video and the features in the face database,and the recognition result is judged by setting the threshold value.This paper build a stable embedded system platform,and test the accuracy of face recognition and feature similarity from face data set and real environment.The experimental results show that the recognition rate of global feature fuse LBP feature is the highest in the illumination environment,the recognition rate of global feature fuse gradient feature is the highest in the occlusion environment,which is 7% and 25% higher than that of the method without local feature fusion,and 12% and 8% higher than that of the embedded face recognition algorithm represented by PCA algorithm in the light and dark environment.The face recognition system designed in this paper can recognize the face identity in different light environment and light occlusion environment,effectively improve the face recognition technology in complex background,the system runs stably and meets the design requirements.