面向监控视频的人脸识别技术及应用

Face Recognition Technology and Application for Video Surveillance

随着深度学习技术的发展,人脸识别在公开人脸数据集上取得了超越人类视觉的进步,具有直观简便、准确可靠、可扩展等优点,在智能视频安防领域受到了越来越多的关注。从实时监控视频人脸识别的研究角度出发,针对监控环境中收集大量样本困难的问题,提出了基于迁移学习的视频人脸特征表达学习网络框架,通过利用海量的网络人脸数据集预训练和视频监控环境下少量样本微调的方法,促进了视频人脸的特征判别能力的表达,证实了该迁移学习对于小样本人脸特征学习的有效性。

With the development of deep learning technology,face recognition has achieved excellent performance on many public face datasets,even better than human vision.For such advantages of face recognition technology as intuitive and simplicity and highly accuracy and reliability and scalability,there are more and more attentions for it in the field of intelligent video surveillance.Aiming at the difficulty in collecting plenty of samples in the monitoring environment,this paper proposes a learning frame network of video face representation expression based on a transfer learning strategy to study the real-time monitoring of video face recognition.By using pre-train of massive web face data and a few sample adjustment in the video monitoring environment,the feature discriminatory ability of the video face is promoted,which proves that the transfer learning is effective for the face feature learning of the small samples.