行人重识别(ReID)
Author:陈维文
行人重识别(ReID)是一项纯粹的计算机视觉 (CV) 任务,其主要目标是在不同的监控摄像头中识别出不同的人物或车辆,并确定它们是否是同一个人或车辆。
监控视频通常难以捕捉高质量的人脸,并且经常会遇到刻意回避或遮盖的情况。为了解决这些问题,ReID 技术应运而生。它可以通过对身体外貌特征和行为模式的分析,对监控视频中的个体进行识别和跟踪。
该任务目前在学术上是检索出不同摄像头下的相同行人图片,同时数据集中只有人的全身照,如下图所示。
但是实际上在实际应用中的时候会和检测结合,简单来说先框出目标后分类,如下图所示。
这个方向做的比较的奇怪,该模块只做整体性介绍,同时希望学习该模块的你对经典网络有所了解。