基于卷积神经网络多方位的人脸检测

Face detection based on convolution neural network in multi-azimuth video surveillance

针对单个摄像头在进行视频监控时视野域有限的问题,提出了一种基于卷积神经网络多方位人脸检测方法。网络的所有权值都是通过学习来不断更新的,所以该网络可以从大量训练集中自动生成特征提取器提取特征,而不需要事先对训练集中的面部图案特征进行手工标记和分析且有很好的鲁棒性。多方位的人脸检测可以更好的捕捉到人脸信息,进而更好地进行人脸检测和识别。同时文中采用改进后的背景差分法,提高运算速率。在通过对人脸数据进行训练的实验结果显示:文中的方法有更高的检测率,并且相对经典卷积神经网络收敛速度更快。

A multi-azimuthal face detection method based on convolution neural network is proposed to solve the problem of limited field of view when a single camera is in video surveillance. The network ownership values are continually updated by learning, so the network can automatically extract feature extraction from a large number of training sets without manual mark and analysis for the facial pattern features in the training set and has a good robustness. Multi-faceted face detection can better capture the face information,and then better face detection and recognition. At the same time it used the improved background difference method to improve the operation rate. The experiments on training with face data show that the method has a higher detection rate and converges faster than classical convolution neural networks.