Design And Implementation Of An Active Video Personnel Monitoring System Based On Deep Learning

Personnel monitoring system is an important part of factory,enterprises and school monitoring systems,and an efficient personnel monitoring system can improve the operation efficiency of the whole factory and enterprise.Because people are the key target of factory and school,the research of the personnel monitoring technology is very important to enhance the function of the video monitoring system in the factory and school.Therefore,the research of the personnel monitoring technology has an important practical value to improve the efficiency of the personnel monitoring and management.The traditional personnel management system is mainly the passive personnel management system based on the identifier or based on human features,and identification and monitoring can be completed with the cooperation from people.Additionally,the functions of these two kinds of personnel monitoring system are limited,and do not realize the function to detect human's pose and the behavior intention.In order to solve the mentioned problems,this article propose an active video monitoring system based on deep learning.The system can combine the personnel monitoring with the security monitoring,use security cameras to obtain monitoring images and realize active personnel monitoring through the intelligent identification software in the background.The main work of this article are as follows:Firstly,a method of face recognition based on deep learning is proposed,which can realize the active identification of people.This article use MTCNN algorithm to detect human face,meanwhile,the algorithm cut and align the detected human face.After that,the FaceNet Network is used to extract facial features.Finally,this article uses k-NN algorithm to realize human face classification.This article uses the Caffe framework to implement the algorithm,and verifies and analyzes the system based on the LFW data set and our own dataset.Secondly,this article studies the human posture estimation method based on the deep learning,and designs the real-time 2D human posture estimation method that is suitable for the factory environment.This article uses the DenseNet network to extract the features of the image,and then design the deep convolution network to predict the joint point and generate the posture information.This article uses the Caffe framework to implement the above method,and test the effectiveness of the system.Thirdly,this article designs and develops the personnel monitoring system software based on the hardware of the video camera.This articles use the GPU acceleration technology and multithread technology to accomplish the software and the Qt to develop the application interface.The function of application software are as follows:personnel recognition,human posture estimation,the display and record of the results of personnel recognition,query of the records and the display of monitoring video.The software is tested in the laboratory environment,and the validity of the software has been proved.