Research And Application Of Low Resolution Face Recognition In Surveillance Video

Video surveillance technology has been widely used in the fields of information security,traffic management and criminal detection.face recognition technology is an important research content in video surveillance.In the real monitoring scene,face recognition based on video image sequence has more flexibility than static images.because the people in the monitoring video are in a moving state and affected by the external environment,the face images captured by camera may be blurred and the resolution is low,which will affect the success rate of face recognition.Therefore,how to improve the accuracy of face recognition under low-resolution conditions is an important research direction,and has certain research and application value.In this paper,we focus on face localization and low resolution image reconstruction in dynamic situations.Specific research content is as follows:1.Aiming at the multi-angles,illumination and occlusion of human faces in surveillance video,the existing face detection methods based on YCbCr complexion model and face detection methods with singular value features are studied.Based on the above,the face detection method of Adaboost algorithm is given.This method mainly uses Harr feature training to construct some weak classifiers,and combines all weak classifiers into one strong classifier to realize face detection.Experiments have verified the effectiveness of the given method.2.Aiming at the problem of low face detection rate caused by illumination change and background interference in outdoor,an improved face detection algorithm combining skin color model and Adaboost algorithm is proposed.The idea of the algorithm is to use the features of the face color brightness and chrominance processing in the skin color model.Before the face detection,the acquired image is treated with light compensation,and then the skin color region is segmented using the skin color model.The area distribution of the skin color is compressed to separate the brightness and chroma of the entire image.Based on this,the face candidate regions are screened,and then the Adaboost algorithm is used to perform the training of the parallel classifier to detect candidate regions and accurately locate the human face.Experimental results show that the improved algorithm improves the facedetection rate under outdoor conditions.3.For the problem of low resolution of the face image obtained in the surveillance video,the process of reconstructing low resolution images to high resolution images is studied.Mainly studied bilinear interpolation,wavelet decomposition and reconstruction,iterative back-projection of the three image reconstruction methods,the former two are for single frame image reconstruction,and iterative back-projection is for multi-frame images with mutual displacement.Refactoring.Based on the analysis and comparison of experiments,for the problem of more jagged textures in the edge part of the iterative back projection algorithm,the image reconstruction method of convex set projection is given,and the high resolution image after reconstruction is well maintained.On the edges and details,the effectiveness of the given method was verified by experiments.4.In terms of face recognition,Gabor wavelet is mainly used to extract facial features.By designing filters of different scales and directions,the real and imaginary parts of Gabor are fitted using the generalized Gaussian distribution,and the real parts of Gabor are The distribution of the imaginary part is used as the texture feature convoluted with the face image,and then the dimension data of the face is reduced by the principal component analysis(PCA).Finally,the face recognition is performed using the nearest neighbor distance discrimination method.The results show the effectiveness of the algorithm used.5.Using the research results of this paper,in the Visual studio2010 environment,with the aid of OpenCV computer vision library,a surveillance video image face recognition system was designed and implemented.Experiments were performed using the actual surveillance video on the campus to achieve the super-resolution reconstruction of people.Face image recognition technology.