Research On Person Re-identification Technology Of Non-overlapping Sights

With the development of society and the improvement of public safety awareness,video surveillance plays an increasingly important role in life.Person re-identification in video surveillance has encountered great challenges due to different camera angles,different backgrounds,pedestrian attitude changes,and occlusion.In this paper,the feature extraction and similarity measure in the non-overlapping horizon descending re-identification technology are studied.In this paper,a feature extraction and fusion algorithm based on overlapping fringes is proposed.Image extraction and overlapping fringe segmentation are performed when extracting person image features.HSV color features,Gabor texture features,color semantic features and person attributes are extracted based on overlapping stripes.Features,and using the typical correlation analysis method to feature layer fusion of the extracted features,effectively solving the effects of illumination,person image background changes and perspective changes.This paper also proposes an algorithm based on contextual content reordering.Firstly,the similarity measure of the extracted features is used to obtain the initial ordering of the person image,and then each image in the initial sorting is reverse sorted,and the initial sorting and inversion are performed.Sorting and reordering the context content to obtain the final pedestrian ranking,so that the recognized correct pedestrian ranking is advanced,and the personrecognition rate is improved.Finally,the proposed algorithm was analyzed and analyzed by using the CMC(Cumulative Matching Characteristic Curve)curve on the three pedestrian re-identification datasets of VIPeR,GRID and CUHK01.Experiments show that the proposed algorithm effectively improves the person recognition rate.In addition,this paper also designs a person re-identification system based on MATLAB image user interface to further verify the effectiveness of the proposed algorithm.