Fault Detection For Video Monitoring System Based On Convolution Neural Network

For a long time,the fault detection framework for video surveillance has been manually viewed by management personnel,this is a huge expense for both time and manpower,and it is difficult to maintain continuous monitoring for 24 hours,faults cannot be discovered and resolved in the first place.So,we will use deep learning to achieve full automatic fault detection of surveillance video.Deep learning is one of the most advanced technologies in the field of artificial intelligence.Convolutional neural networks have a significant effect on image processing in deep learning models.Compared with the traditional neural network,the convolutional neural network can automatically extract and learn the image features,and provide accurate classification and prediction.In addition,it has good robustness for scaling,stretching,and distortion of the image.This thesis first describes the current development of video surveillance,system structure,working principles and characteristics of each part in detail,The most common faults in surveillance video are classified according to the phenomenon.The characteristics and causes of various fault conditions are elaborated in detail,and a variety of fault handling methods are listed.Subsequently,the means of image quality assessment was introduced.According to the characteristics of the video surveillance system and the evaluation criteria,a set of new standards for the judgment of video surveillance faults was constructed.The video fault diagnosis is realized through a convolutional neural network model.Afterwards,this thesis optimizes the basic convolutional neural network model,and re-selects the activation function and cost function of the network.The adjusted network is more adapted to the characteristics of the surveillance video and can accurately classify the common faults of the surveillance video.Finally,the video surveillance fault detection system be designed based on the above content.The system has a two-stage structure.Based on the method of image quality assessment in the first stage,the image is judged whether there is a fault,and the possibility of missed judgment is minimized.In the second stage,a multi-category convolutional neural network is used for the fault image.The fault condition is identified.This system is different from other fault detection systems,avoiding missed judgments caused by failure to identify fault features,greatly improving the detection range and accuracy,and according to the characteristics of convolutional neural networks,can be optimized after used,end expand the scope of model recognition.The system was embedded into the security monitoring system for trial operation.Through the analysis of test results generated in the test run,summed up the current system of basic performance and deficiencies,and clear the future direction of improvement.