Research And Implementaiton Of Safe State Detection Based On Deep Learning

As a tool for the preservation of important materials,safe is also used to ensure the safety of property.In the past,the anti-theft technology of safe was mostly placed on the development and revision of locks,such as fire-proof safe,fingerprint identification safe,etc.,but relying solely on the anti-theft measures of the safe itself,the safety of the safe cannot be guaranteed.With the development of deep learning technology and the implementation of the national “AI+X” strategy,the intelligent monitoring system based on deep convolutional neural network shows great application prospects.Therefore,based on the surveillance video data of safe,the research is based on deep convolutional neural network and computer vision technology safe target switch state detection algorithm,and realize the safe state detection system of safe switch.The main work of this paper is as follows:First,For the problem of safe switch state detection,this paper selects 50 hours of real video from four different scene cameras,and selects a total of 14705 images with low similarity as the target detection data set.It covers the actual environment of different camera resolutions,different lighting environments and safe locations.Second,Two safe state detection methods are implemented based on the safe target detection data set.One is to modify the general target detection algorithm SSD,Faster RCNN and RFCN feature extraction network,directly detect and classify the safe,and finally compare the performance of each general algorithm on the data set;The other is to classify the safe after detect the location of the safe,By designing a classification network of safe switch states,the detected safe targets are reclassified.Third,In view of the characteristics that the body posture will change when the person switches the safe,exploring the position change of the human key point when the person switches the safe,and using AlphaPose algorithm to apply human body key point detection network to the safe state switch detection.The detection accuracy of the safe state switch is improved by combining with the target detection algorithm.Fourth,designing and achieving the intelligent analysis system of safe switch state detection based on deep learning.The system can detect and visualize the state of multi-channel video safe in real time,and can judge the event of the switch safe and display the historical event.