Joint Gaussian And Semantic Feature On Person Re-identification

With the stable and harmonious development of society,more and more people began to pay attention to social security issues.Intelligent surveillance system has always been an important platform for people to find dangerous situations and timely troubleshooting.Person re-identification is an emerging research direction in the field of computer vision,and it is also an important means of intelligent security detection system.Person re-identification is mainly to extract the feature map for person,to determine whether the two images indicate the same person.Therefore,this paper aims to study how to improve the discrimination of the features.In spite of the seemingly simple task of re-identification,there is a great challenge to re-identification tasks because of the fact that two images are taken in a variety of environments,such as different times,different locations,and different camera views.We find that the semantic features from deep learning can not be better focused on special peculiarities of person.However,the Gaussian feature obtained by traditional machine learning can describe the color,texture and relative position of the body and other information for person,and has rich person characteristics.Therefore,this paper combines traditional Gaussian features with deep semantic features to enhance the discrimination of the overall features of person.In this paper,we describe in detail the framework of person re-identification and fusion strategies,and we have achieved good performance on two public data sets(Market1501and VIPeR).In addition,we apply this model to the vehicle re-identification,which combined with person re-identification in the method.Experiments show that our method in the vehicle data set VeRi achieved good results.This also illustrates the interrelationship of re-identification tasks.We compared the results with the current high level of results,and further proved the effectiveness of our method.