Design And Implementation Of Early Warning Of Intrusion System For Construction Machinery In Risk Area Of Oil And Gas Pipeline Based On Edge Computing

By the end of 2018,the usage of oil and gas accounted for 57.48% of the global energy market.With the pipeline transportation’s advantages of safety,economy and large transportation volume,makes pipelines laid more and more widely.In recent years,with the development of urban and rural construction,accidents of third-party damage to pipelines have occurred frequently,which not only causes significant economic and property losses,but also causes great harm to personal and public environmental safety.Therefore,pipe network companies often dispatch inspectors,but many high-consequence area pipelines are located in remote areas,making manual inspections difficult and heavy workload.Also in third-party construction sites,traditional video surveillance is usually used,which cannot detect and identify large construction machinery that may cause damage to pipelines.It can only be viewed by people,which is easy to have false negatives and false positives,and the efficiency is not high.Based on this,the research on the early warning of intrusion system for construction machinery in risk area of oil and gas pipeline was carried out.The main work of this paper includes the following three aspects:1.Aiming at the risk areas such as third-party construction sites of pipelines and high-consequence areas,an early warning of intrusion system for construction machinery in risk area of oil and gas pipeline based on edge computing was designed and developed.The system is divided into two parts: edge surveillance devices and the pipe network dispatching center.The edge surveillance device is placed on third-party construction sites and high-consequence areas.At the edge,device uses image processing technology to complete the detection and recognition of large construction machinery to obtain two-dimensional coordinates,and then determines whether the two-dimensional coordinates collide with the risk area.Once a collision occurs,it is determined as mechanical invasion.The alarm will be pushed to the dispatching center.The dispatching center receives and views alarms from the edge surveillance device.It also supports configuring the edge surveillance device.2.Analyzed and compared the performance of various object detection algorithms on the embedded computing platform Jetson Nano,and finally Mobile Net V1-SSD is selected as the object detection algorithm of the system.Meanwhile,the accuracy of the model was improved from three aspects of data enhancement,SE model addition and Anchor optimization.Compared with the base model,m AP@0.5 improved by 16.3%.3.In order to solve the instability of the object detection algorithm based on still image,the tracking effects of TLD,BOOSTING,KCF three object tracking algorithms in the actual application scene are analyzed and compared,and finally KCF is selected as the object tracking algorithm for the system.And a re-detection tracking algorithm is proposed to alleviate the drift problem.The field test shows that the system can identify large machinery in real time and accurately,the precision is 92%,the processing speed is stable at 25 fps,which meets the requirements of cooperative enterprises,and has a certain engineering practical value.