Person Re-identification Algorithm Combining Visual Features With Mapping Model Learning

In recent years,person re-identification has received growing attention with the increasing popularity of intelligent video surveillance and the advancement of safety sense.It aims at spotting a person of interest in other cameras.This paper will raise three new person re-identification algorithms based on the analysis and summary of the domestic and international relevant research results.Firstly,aiming at the pedestrian-wise discrepancy and the view-wise discrepancy across non-overlapping camera views,we address the visual feature extraction and cross-view mapping model learning.In the order of pedestrian similarity,we assign the different pedestrian samples by iterative reweighted sparse representation,which can make different weights to suppress the outliers and thus get a robust matching model.Secondly,combining local maximal occurrence feature and the cross-view feature mapping model learning,the person re-recognition algorithm research the change of the illumination of non-overlapping camera views and the completely discriminative features extraction.First of all,the local maximal occurrence feature is extracted from the person image,so as to overcome the change of illumination and extract completely discriminative features.Then,we employ the learned mapping model of the cross-view to transform features hence eliminate the different features of the camera views.In the end perform distance measurement and sorting.Finally,in consideration of background interference problem in the person image,to begin with we formulate deep decompositional network to accurately and availably extract foreground of person image thus achieving the purpose of suppressing background interference problem.And then extract the visual features of images.At last learn the cross-view feature mapping model for feature transformation,distance measurement and sorting.