CurtinFaces Database

This database was captured at the Department of Computing in 2011. The speciality is that the data is captured using a Microsoft Kinect Sensor. Unlike traditional 3D scanners, the Microsoft Kinect trades data quality for low cost and high speed, which is usually a requirement for broad practical applications. A total of over 5000 samples are captured from 52 subjects randomly selected from our department including a mix of male/female and glasses/without glasses. These images have varying facial expression, view-point, illumination and occlusion, simulating a real-world uncontrolled face recognition problem. For each Microsoft Kinect capture, a standard Panasonic digital camera is also used to capture the face image at the same time. The 3D point cloud and RGB data are stored in Matlab (.mat) format.

本数据集于2011年在计算机系采集完成。该数据集的采集特色为采用微软Kinect传感器（Microsoft Kinect Sensor）获取数据。与传统三维扫描仪不同，微软Kinect传感器以牺牲部分数据质量为代价，实现了低成本与高采集速度的优势，这一特性恰好契合大规模实际应用的普遍需求。本次采集共从本系随机选取的52名受试者中获得超过5000条样本，受试者涵盖男女、是否佩戴眼镜等不同群体。所有样本包含多样化的面部表情、拍摄视角、光照条件与遮挡情况，模拟了真实无约束场景下的人脸识别任务。在每次通过微软Kinect传感器采集数据的同时，还会借助一台标准松下（Panasonic）数码相机同步采集面部图像。三维点云与RGB数据均以Matlab（.mat）格式存储。

搜集汇总

背景与挑战

背景概述

CurtinFaces Database是一个2011年采集的人脸识别数据集，使用低成本、高速的Microsoft Kinect传感器捕获，包含52名受试者的超过5000个样本，涵盖多种面部变化和真实世界场景，数据以Matlab格式存储3D点云和RGB图像，适用于计算机视觉和人工智能研究。