Deep Learning Approach to Face Pose Estimation for High-Speed Camera Network System
Deep Learning Approach to Face Pose Estimation for High-Speed Camera Network System
复制标题
高速摄像网络系统人脸姿态估计的深度学习方法
DOI：
10.1109/icaiic48513.2020.9065051
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
M. Ishikawa
中科院分区：
文献类型：
--
作者：
Seohyun Lee;Hyuno Kim;M. Ishikawa
三维人脸姿态估计在计算机视觉中得到了广泛的研究，因为人脸识别技术不仅可以用于人类行为监测，而且可以用于人机交互。在本文中，我们尝试建立一个深度学习模型，该模型通过直接应用卷积神经网络来分类人类头部的摇摄角度，而无需进行初步的图像处理，用于低分辨率的人脸图像。与基于预训练模型的迁移学习相比，由几个卷积层和dropout方案组成的定制简单模型在人脸平移角度预测方面表现出更高的准确性。
Three-dimensional face pose estimation has been vastly researched in computer vision, as the face recognition techniques can be utilized in tremendous applications not only regarding human behavior monitoring but also about human-computer interaction. In this paper, we attempted to build a deep-learning model which classifies the pan angle of human head by directly applying convolutional neural network without preliminary image processing, for low-resolution face images. In comparison with the transfer learnings based on pre-trained model, customized simple model consisting of a few convolutional layers and dropout scheme showed an enhanced accuracy in face pan angle prediction.