研究目的
To propose a face recognition method that utilizes both depth and infrared pictures to overcome the limitations of conventional color picture-based methods, such as susceptibility to illumination changes and vulnerability to fake face attempts.
研究成果
The proposed method effectively recognizes faces in both normal and low illumination environments by utilizing depth and infrared pictures. It addresses the limitations of conventional color-based methods but requires further study to improve accuracy for rotated faces.
研究不足
The accuracy decreases when the face is rotated beyond 30 degrees. The method requires compensation for depth value changes due to face rotation.
1:Experimental Design and Method Selection:
The method involves face detection using depth pictures for speed and feature extraction from infrared pictures using 3D Local Binary Pattern (3D-LBP) for accuracy.
2:Sample Selection and Data Sources:
Captured images of 20 individuals under various lighting conditions and poses.
3:List of Experimental Equipment and Materials:
Kinect v2 for capturing depth and infrared pictures.
4:Experimental Procedures and Operational Workflow:
Detection of nose points in depth pictures, setting ROI for face detection, feature extraction using 3D-LBP, and face identification by comparing histograms of feature pictures.
5:Data Analysis Methods:
Chi-square distance for histogram comparison to measure face similarity.
独家科研数据包,助您复现前沿成果,加速创新突破
获取完整内容