融合行人属性特征的ReID重排序方法

ReID Reordering Method Integrating Person Attribute Features

在行人重识别问题中,为了减少因光线、摄像头角度、遮挡等因素对搜索结果带来的影响,需要通过拓展维度来丰富图像特征从而提高搜索精度。为此,本文提出了一种对行人重识别搜索结果进行重排序的方法。该方法首先通过行人的全局特征对行人图像进行搜索排序,然后提取行人属性特征,并利用属性特征训练一个二分类模型。通过模型对搜索结果进行重排序,并评估搜索精度。实验的结果表明,在加入属性特征之后,行人重识别算法搜索的Rank-1、mAP值均有提升。

In pedestrian re identification problems,in order to reduce the impact of factors such as lighting,camera angle,and occlusion on search results,it is necessary to enrich image features by expanding dimensions to improve search accuracy.Therefore,this article proposes a method for reordering the search results of pedestrian re identification.This method first searches and sorts pedestrian images based on their global features,then extracts pedestrian attribute features and trains a binary classification model using these features.Reorder search results through a model and evaluate search accuracy.The experimental results show that after adding attribute features,the Rank-1 and mAP values searched by the pedestrian re identification algorithm are both improved.