节点文献
基于计算机视觉的车间人员增强检测技术
【作者】 刘洋；
【导师】 徐骏善；
【作者基本信息】 南京理工大学 ， 机械电子工程， 2021， 硕士
【摘要】 车间是制造活动执行的基础环境,由于车间人员有意志自主性,其运动速度、状态、方式等难以表征,因此车间人员管控是车间生产作业中最难控制的环节,军工产品线更是如此。由于军工品产品的特殊性,往往对生产作业过程中的不同区域的人数、不同作业环节与工序等的人员位置有着严格要求。一旦危险发生,人员聚集过密将造成灾难性安全事故,严重威胁车间生产人员人身安全。因此,如果可以对该类车间生产作业人员进行实时准确管控,则既可保障安全生产,又可提高生产效率。对此,本文分析车间真实生产需求,构建系统功能模块和整体框架,提出一种基于计算机视觉的车间人员增强检测的智能管控方法。本文采用车间人员检测技术,提出自适应识别网络,实现车间中人员的视觉化智能检测;针对自适应网络识别失败的问题,基于车间人员运动学模型对车间人员进行跟踪,预测消失目标的理想出现位置,补全漏检以实现检测结果的二次增强;对于多源数据融合问题,使用地理围栏技术,实现重叠区域重复识别结果的删除;使用基于像素坐标系到世界坐标系的映射矩阵,实现车间多区域人员定位。最后基于Web设计车间人员增强检测及定位系统,实现车间全局内人员数量及位置的实时显示。目前该系统已交付航天集团某厂实际使用,结果表明该系统大大降低了安全监管对管理人员的依赖性,成功实现车间人员智能管控。
【Abstract】 Workshop is the basic environment for the execution of manufacturing activities,and the staff have the characteristics of subjective initiative and movement uncertainty.The supervision of the staff has always been the most difficult part of the production activities,especially the military production lines,due to the particularity of which there are often strict requirements on the number and density of person in different areas,different operations,and different production processes in the production process.It may cause catastrophic safety-accidents in the local areas where the density of staff is too much,which threatens the safety of the staff seriously.Therefore,if real-time and accurate control of production workers in this type of workshop can be carried out,safe production can be guaranteed,so as production efficiency.This thesis takes real needs and technical difficulties of the workshop into consideration,builds the system architecture and functional modules.We propose an enhanced detection technology for staff management based on computer vision.We use workshop staff detection technology,an adaptive recognition network is proposed to detect staff smartly in the workshop.When adaptive rec-network cannot recognize the object successfully,kinematic modeling is proposed to predict the position of the disappearing target,so as to complement the missed detection and enhance the detection result.Geo-fencing is used to make multi-camera data fusion come to real,thus correcting the repeated results.The real coordinates of the staff in the workshop can be computed by figuring out the mapping relationship between the pixelcoordinate-system and the world-coordinate-system.Based on the technology above,the number and location information of the staff in the workshop can be calculated all-around.Finally,the results are synchronously displayed on the Web system.The system has been delivered to the enterprise for actual use,which greatly reduces the dependence on management staff,and successfully realizes the intelligent management of workshop staff.