Person Reidentification Based On Deep Nerual Network

With the development of intelligent surveillance,the practical value and theoretical significance of person reidentification in the field of intelligent video surveillance are increasingly prominent.Person reidentification is a technology to determine whether there is a given target in the image library or video sequence using computer vision technology in the data set collected under surveillance camera.With the improvement of computer hardware performance,the research of person reidentification based on deep neural network has aroused the interest of researchers and become a hot topic of computer vision.This research involves many research fields such as image processing,computer vision,machine learning,image retrieval,etc.The theory has important scientific significance and can be widely applied in the application fields of computer vision,such as intelligent security and security etc,with a good application prospect.In practical application,person images are faced with many problems,such as low resolution,large differences in human posture and camera angles,and obstruction,which will lead to large differences in human appearance.To solve these problems lead to the low quality of image and image difference problem in sequence.In this paper,considering the interframe abundant temporal information,on the basis of quality awareness network,this paper proposes a supervised temporal attention quality awareness network,by extracting timing information between frames,combines the single frame image space features and characteristics of the movement between frame and frame,thus effectively aggregation between all the frames the complementary information,significantly reduce the effects of low quality image area,improves the robustness of low quality image.However,the supervised method requires a large amount of annotated training data,which is difficult to realize in the real world.Aiming at this problem,this paper further puts forward a model based on the depth of unsupervised clustering network of pedestrian recognition method,this method is based on across the diversity between the camera and the camera inside the similarity,will begin a single sample as a different identity,the introduction of a variety of regularization item to balance the amount of data,each clustering to maximize each sign the diversity of clustering,and gradually form a similar sample identity ID clustering to merge,realize the diversity and similarity of effective balance.In order to verify the effectiveness of the method proposed in this paper,experiments were carried out on four public data sets(PRID 2011,iLIDs-vid,CUHK03 and Market1501)and the latest relevant algorithms.The experimental results show that the method proposed in this paper has a good effect on the recognition of matching rate and computational efficiency.