Cross-camera Person Re-identification And Retrieval System

With the concept of "Internet Plus initiatives" proposed,building a more harmonious and safer social environment has gradually become the main appeal of people's security construction at this stage.In recent years,the rapid development of deep learning and artificial intelligence has provided new development directions for the video surveillance system.The problem of cross-domain person re-identification in video surveillance system has attracted a large number of scholars' research in recent years.However,due to the quality of images captured by cameras,the illumination intensity,the viewing angle and the change of the pedestrian posture,the appearance characteristics of the same pedestrian in different monitoring videos will be greatly different.Based on the above problems,this thesis designs a feature fusion and ellipse segmentation ReID algorithm from the aspects of design features and image segmentation.In addition,in order to apply to the actual system,this thesis uses the feature-based word bag model to speed up the retrieval.The main work and innovations of the thesis are as follows:1.In the feature part,this thesis firstly uses the combination framework of global and local features.This structure can guarantee the integrity of feature information and highlight local feature information.Then,LOMO and GOG are used as composite features with several basic features as global feature,combined with local HSV and LAB color features,can fully exploit the complementary characteristics of LOMO and GOG features,thus forming a more robust feature representation.2.The segmentation part adopts elliptical segmentation and horizontal uniform segmentation.By observing images on the person re-identification datasets,we can find that most of the four right-angled regions are useless interference information.The elliptical segmentation can remove some unrelated background interference and reduce the computational complexity.In order to highlight the local details,this thesis also uses a horizontal uniform segmentation method to increase the details of the image and the effectiveness of the feature.3.Combining the above work and algorithm characteristics,a cross-camera person re-identification retrieval system based on B/S framework is designed.In this system,a fast retrieval algorithm based on the word bag model is added to improve efficiency of the system retrieval,thus realizing the rapid retrieval of person.