Research And Design Of Intelligent Elevator Monitoring System Based On Deep Learning

More high-rise buildings has brought about a rapid growth in the number of elevators,which brings great convenience to people,but also created many safety problems.The way of installing surveillance cameras in the car plus manual monitoring in the monitoring room 24 hours a day is adopted for ensuring the safety of elevator passengers.Manual monitoring has alleviated the safety problems of elevators to a certain extent,but there are still problems such as low efficiency and untimely detection of abnormalities.With the support of the Industrial Upgrading Project of Jiangsu Provincial Department of Industry and Suzhou Key Industrial Technology Innovation Program,the elevator intelligent monitoring system is researched and implemented.The system includes a deep learning-based car forbidden target detection algorithm,a door blocking behavior recognition algorithm and the design and implementation of a smart elevator cloud platform.The intelligent monitoring system is of great significance to ensure the safety of elevator passengers and improve the quality of elevator services.First,according to the demand of intelligent elevator monitoring,the research and development objectives are determined and the overall architecture of the system is developed.According to the system architecture,the problems of detecting the car forbidden entry target and detecting the behavior of passengers blocking the door is analyzed,main methods to be used is determined as well.Secondly,image-based detection algorithms for prohibited car entry targets are investigated.The research goal is to achieve timely detection of electric vehicles that are forbidden to enter the car in violation.Firstly,the elevator surveillance video is pre-processed to filter out frames with moving targets.Then the Efficient Det network is improved based on the cross stage partial network to achieve a more lightweight framework and to improve the detection speed.Tests show that the improved Efficient Det network achieves higher detection efficiency.Again,for the demand of intelligent detection of passenger door blocking behavior through elevator surveillance video,a behavior detection method based on generating candidate video segments and 2D+3D convolution is proposed.The method is divided into three steps,the first step is to generate candidate video segments with sliding windows.In the second step,the behavior in the video segment is recognized using a behavior recognition algorithm combining2 D convolution and 3D convolution.Finally,a non-local neural network is introduced to improve the global information extraction capability of 3D convolution.Tests show that the method can meet the online detection accuracy of door blocking behaviors.Finally,the intelligent elevator monitoring cloud platform system is designed and implemented,including the server side,database and client side.The server side can realize the automatic collection of elevator operation data and monitoring video.The database realizes the storage of video and picture data,and the client visualizes the operation data,monitoring video and its intelligent detection results.The server side,database and client side of the cloud platform system work together to realize the intelligent cloud monitoring of elevators.