Research And Application Of Violent Event Description In Video Based On Deep Learning

In the future "smart city",The monitoring camera will be placed in every corner of the public place.These monitoring devices will produce huge amounts of video and related data.Detection and description of events in the monitoring video.These can produce great application value in urban public safety,traffic management and so on.Especially in the surveillance video under the space of bank ATM and elevator,there are frequent violence in society.If we can get safety alerts in time,we can greatly reduce personal and property losses.But video in a public scene is characterized by a low density of information and a large scale of data.In addition to the objective problems such as complex and diverse background scenes in video,the monitoring video analysis in public places is quite difficult.In view of the above problems,we choose the data set and the depth network model which is most suitable for the description of the video violence,and make the integration and improvement.It meets the needs of the computer to quickly and accurately describe and alert the violent events in the video.First,on the basis of the research results of existing deep learning,this paper chooses the convolution neural network and recurrent neural network.The image frame is decomposed from the monitoring video,and then processed by pixel level recognition.Combined with video and text data,the depth model corresponding to the corresponding video frames can be trained.Secondly,This paper focuses on the problems encountered in the application of deep learning methods to the engineering field,such as operation efficiency,accurate description and so on.The system optimization is realized through image preprocessing,micro tuning network structure,context awareness,coarse and fine granularity combination processing and so on.In this paper,a more advanced and deep learning method is applied to analyze the video and build a high availability and extensible description system for violent incidents.This paper compares the advantages and disadvantages of different convolutional neural network models,takes account of the detection accuracy and speed,designs the most suitable network model for each part,and optimizes network parameters in training process.The proposed algorithm is superior to other existing algorithms in the detection precision,and the detection speed meets the requirements of the actual application scene.