基于深度学习的远程视频监控异常图像检测

针对复杂场景下远程视频监控图像异常检测困难、传统算法功能单一(仅针对某种特定场景或某种异常图像进行检测)等问题，提出一种基于深度学习的全自动远程视频异常图像检测方法。首先采用Xavier方法对自行设计的卷积神经网络(Convolutional Neural Network,CNN)的参数进行初始化，然后将标准化后的视频差分图送入CNN的输入层，通过特征提取及下采样，最后在CNN的输出层获得远程视频异常图像检测结果。实验结果表明，该方法可以对远程视频监控中突然出现遮挡、模糊和场景切换等多种异常同时进行实时在线检测，准确率可达88.75〖WT《Times New Roman》〗%〖WTBZ〗。

Detection of Abnormal Remote Video Surveillance Image Based on Deep Learning

It is difficult to detect abnormal image in complex scene during remote video surveillance and the function of traditional method is single(lonly for a specific context or a specific abnormal image),a deep learning based full-automatic method is proposed to detect remote video abnormal images.Firstly,Xavier is adopted to initialize the parameters of the self-designed convolutional neural network(CNN).Then normalized video differential images are sent to the input layer of CNN.Finally,by means of feature extraction and downsampling,results for abnormal images detection of remote video can be obtained in the output layer of CNN.The experimental results show that the proposed method can conduct real-time online detection of various abnormal images such as image occlusion,blurring and scene switching in the remote video,and the accuracy rate is up to 88.75%.