Research And Application Of Intelligent Monitoring System Based On Deep Neural Network

Nowadays,the security situation of society is becoming more and more complicated,and the effect of traditional monitoring and recognition algorithm can't meet the demand.With the rapid development of computer vision technology and deep learning theory,intelligent algorithm can replace the traditional recognition algorithm.This thesis mainly considers the recognition algorithm based on depth monitoring network,and applied the algorithm in intelligent monitoring system.Different from the traditional monitoring and recognition algorithm,the depth network in this thesis does not need to design the feature artificially,and it is not necessary to adjust the parameters artificially according to the environment and algorithm.This thesis designs a multi-scale convolution neural network to train the network model that can extract human behavior in the behavior video library and use it as an algorithm to monitor abnormal behavior.In addition,based on the YOLO network,the pedestrian image library is used to optimize the feature extraction ability of the network.The intrusion detection algorithm and the wandering detection algorithm are designed.The algorithm has achieved excellent recognition effect in the simulation experiment.Finally,the algorithm is applied In the intelligent monitoring and identification system.This thesis focuses on the implementation of improved multi-scale neural network behavior recognition algorithm and YOLO-based intrusion detection algorithm and wandering detection algorithm,the specific work is as follows:1.The 3D CNN is studied,and the input layer is improved by longer video input and the parallel convolution of the gray channel and the optical channel,so that the convolution kernel of the first layer in the convolution layer can be focused on the feature extraction.In the network structure,the three-dimensional space-time subsampling layer is added to the network to increase the translation invariance on the time scale,and the robustness of the network to the video recognition is improved.The NIN structure is added to make the network have the non-linear feature extraction ability.Space-time pyramid structure is added,making the whole network can enter different resolution,different length of video,making this model applies to the real environment of video surveillance,finally strengthen the application value of the network.2.The YOLO network is trained to strengthen the human body feature extraction ability in the network,and the network is combined with the traditional intrusion detection algorithm as a feature extractor to realize a new intrusion detection algorithm.On the basis of this intrusion detection algorithm,by combining the motion direction characteristics and color characteristics,a hover detection algorithm which can automatically detect the human body and can resist the occlusion is realized.This algorithm achieves a higher detection effect than other similar algorithms.Compared with the traditional intrusion detection algorithm,this algorithm can distinguish whether the intrusion is the human body,with greater practical value.3.According to the requirement analysis of real intelligent monitoring system,the whole framework of the server is designed.The human behavior recognition algorithm based on three-dimensional multi-scale convolution neural network,and the intrusion detection algorithm and wandering detection algorithm based on YOLO model,which are original algorithms of this paper,are applied to the identification service of intelligent monitoring system.Through accelerated optimization,the system finally achieved the ability to support more than 16 real-time identification services.