Research On Pedestrian Re-identification Algorithm Based On Deep Learning

With the development of the city and the progress of science and technology,the urban public infrastructure is becoming more and more perfect,especially the layout of "urban skyeye system",which makes the main traffic roads and streets and lanes in the city full of cameras.The layout of "Eye of Heaven" has greatly improved public safety to a certain extent.The problem of Person Re-Identification is to find all the images of the same pedestrian from a huge pedestrian image database as accurately as possible.The development of Person ReIdentification technology has great theoretical significance and practical application value in the field of security monitoring in smart cities.Video images captured by urban cameras,especially in crowded scenes,target pedestrians are vulnerable to occlusion,camera angle,illumination changes,background changes and other factors,which will seriously affect the accuracy and robustness of Person Re-Identification algorithm.The research on Person ReIdentification can be summarized in two aspects:(1)acquiring features that can accurately describe image information.(2)building a mechanism based on image features to accurately measure image features,so that it can efficiently and accurately retrieve images consistent with the query image.In this paper,the main work is as follows:(1)Using convolution neural network to obtain image depth features as image descriptor,a classification-discriminant model based on Siamese network is used to search and match pedestrian images.(2)Person Re-Identification open data sets are limited in scale.The Generative Adversarial Networks is used to generate pedestrian images with different styles to expand the data set.The expanded data set is used to train the Person Re-Identification and improve the learning ability of the model for pedestrian features.(3)A re-ranking algorithm for identification results is proposed.By calculating the distance between the query image and the result images.The pedestrian image closest to the query image is placed in the front of the result image sequence,that can improve the performance of Cumulative Matching Characteristic and mean Average Precision of the model.