【作者】 李凡；

【导师】 孙长银； 牛丹；

【作者基本信息】 东南大学 ， 模式识别与智能系统， 2020， 硕士

【摘要】 随着城市建设用地的日益紧缺以及汽车保有量的大幅增加,立体车库得到了大力发展,与此同时也带来了很多运维管理问题。目前的立体车库多未联网,采用人工值守监管的方式,这种方式不仅费时费力,而且智能化水平较低。针对这一现状,在江苏省重点研发计划项目的支持下,本文研发实现一套立体车库云端智能运维系统,包括立体车库电机故障诊断算法、基于监控视频的智能行人检测算法以及云端智能运维系统的服务端和客户端设计与实现,大幅提升了立体车库运行的安全可靠性、监管效率和整体智能化水平。首先,分析立体车库运维管理的业务特点,确定各服务功能模块,将智能运维的核心算法与云平台开发技术相结合,基于阿里云设计立体车库云端智能运维系统架构与总体方案。其次,针对立体车库齿轮减速电机的故障诊断需求,对电机振动信号进行时域与频域分析,采用DBN提取振动信号时频域指标的深度特征,并根据电机样本数据的正负样本悬殊的特点,设计和实现了基于DBN-OCSVM的车库电机故障诊断模型。通过与PCA特征提取算法的对比实验,验证了该模型在立体车库齿轮减速电机故障诊断的准确性,提高了立体车库运行的安全性。然后,针对立体车库的行人目标检测需求,提出一种基于帧差法与YOLOv3的立体车库行人检测算法,提高行人检测的检测效率,同时减轻云端服务器的计算压力。首先使用帧差法筛选出存在运动目标的视频片段,再针对筛选后的视频使用YOLOv3进行行人检测。通过与目前常用的目标检测算法的对比实验,验证了该立体车库行人检测方法的高效性与准确性,显著提高了无人值守式立体车库的可靠性与管理效率。最后,融合立体车库电机故障诊断算法与基于视频的行人目标检测算法,设计和实现了立体车库云端智能运维服务端和客户端系统,包括车库总览、车库管理、实时监控、故障管理、系统管理、统计报表以及维保服务七大功能模块。系统在项目合作方进行了现场联调与实际运行测试,测试结果表明本论文研发的立体车库云端智能运维系统显著提升了立体车库智能化和网络化运维管理水平。更多还原

【Abstract】 With the increasing shortage of land for urban construction and the substantial increase in car ownership,the stereo garage has been developed rapidly,and many problems of operation maintenance have become increasingly serious.At present,most of the stereo garages are not connected to the Internet,using manual supervision,which is time-consuming and laborintensive,and the level of intelligence is low.In response to this situation,with the support of Primary Research and Developement Project of Jiangsu Province,a cloud Intelligent operation and maintenance system for stereo garage is developed in this thesis,including the development of fault diagnosis algorithms of garage motor,intelligent pedestrian detection algorithms based on video,and the realization of the server and client of the cloud intelligent operation and maintenance system,which improves the safety and reliability of the stereo garage,management efficiency and the level of intelligence.First,the needs of the stereo garage operation and maintenance system are analyzed in this thesis,the module of each service is determined,key algorithms and the development technology of cloud platform are combined,and the architecture and overall plan of the cloud intelligent operation and maintenance system for stereo garage are designed.Secondly,according to the fault diagnosis requirements of the stereo garage motor,the vibration signal of the motor is analyzed in the time and frequency domain,and then DBN is used to extract the depth characteristics of the vibration signal in the time and frequency domain.According to the characteristics of the positive and negative motor samples,the motor fault diagnosis model based on DBN-OCSVM is designed and implemented.Through the comparison experiment with PCA feature extraction algorithm,the accuracy of the model in the motor fault diagnosis for stereo garage is verified,and the operation safety of the stereo garage is improved.Then,according to the needs of pedestrian detection in the stereo garage,a pedestrian detection algorithm based on the frame difference method and YOLOv3 is proposed to improve the detection efficiency of pedestrian detection and reduce the computing pressure of the cloud server.First,the frame difference method is used to filter out the video clips with moving targets,and then the YOLOv3 pedestrian detection model is used in the filtered video.Through the comparison experiment with some pedestrian detection algorithms which are current used commonly,the efficiency and accuracy of the pedestrian detection method for the stereo garage are verified,and the reliability and management efficiency of the stereo garage are significantly improved.Finally,the cloud intelligent operation and maintenance server and client system for stereo garage are designed and implemented through the fusion of stereo garage motor fault diagnosis algorithm and video-based pedestrian detection algorithm,including garage overview,garage management,real-time monitoring,fault management,system management,statistical reports and maintenance services.System joint debugging and actual operation testing were carried out in the project partner company.The test results show that the cloud intelligent operation and maintenance system for stereo garage developed in this thesis significantly improves the level of intelligent management in the stereo garage.更多还原