Research On Person Re-identification Algorithm In Non-overlapping Views

In recent years,with the growing of the high attention to the public security and the development of video surveillance system,the intelligent video surveillance system gets a lot of popularity.Person Re-identification is a new technology inInteligent video in recent years and it is also one of the key issues which is need to be solved in the multiple cameras intelligent video monitoring systems and also get the attention of the research in computer vision and artificial intelligence research.Person re-identification in non-overlapping views is a process that we need to judge a people in a camera whether or not appers in another camera which is in non-overlapping views at the cameras video network monitor system environment,it is also to identify all the pictures of the same people captured by the camera.But person re-identification is a very challenging problem because of that the same person may have a big difference in different camera environments for the impact of the many external factors that camera angle,background change,light conditions,attitude change,object shelter and so on.In this paper,we puts forward a metric learning of person re-identification algorithm RPML which is mainly identify a person again by generating a measure matrix of characteris learning.This paper first make a deal with the original person image through an image enhancement algorithm to reduce the effects of the illumination changes.After that we make a reasonable segmentation for the person according to the human body target shape symmetry and extract the pedestrian image color features(HSV,Lab),texture features(FHOG,SILTP),ColorNames features and change the original linear feature space to a more distinct nonlinear characteristic of the space with a PCA dimension reduction by a kernel function learning.We learn three separate measure matrix and optimize the measure metric overfitting problem because of the difference of different types of features of pedestrian image description.Finally we get the similarity measure function of the pedestrian image by a weighted fusion of multiple measure metric and implement a function of the similarity measure of the people.Lastly we make a experiment and make a compare by CMC(Cumulative Matching Characteristic Curve)cure on three pedestrian data sets,VIPeR,ILIDS,CHUK01.