| The emergency assessment of post-earthquake building safety is to determine whether the building is suitable for continued use after the earthquake and whether it has the ability to cope with aftershocks.Previous experience in earthquake damage assessment shows that,on the one hand,the post-earthquake environment is complex,many earthquake damage locations cannot be directly observed,and it is difficult to quickly complete all assessment tasks by relying on the manual operation of limited professionals.On the other hand,the inconsistency of standardized workflow and evaluation evaluation standards has caused the assessment results to rely heavily on the professional quality of personnel,resulting in doubts about their accuracy.In this paper,the identification,damage identification,failure pattern recognition and dynamic deformation monitoring of reinforced concrete(RC)frame structure members are analyzed and studied by computer vision technology and digital image processing technology,so as to provide a basis for the post-earthquake safety emergency assessment of building structures.The relevant research in this article is as follows:(1)The RC frame structural exponent identification model was constructed based on Mask R-CNN,and two independent databases including beams,columns and nodes were sorted and screened using actual seismic sites and laboratory specimens as data sources,and the labeling methods were determined according to the morphological characteristics of beams,columns and nodes,and the generalization ability of the component recognition model on small data sets was improved through data enhancement methods such as image transformation and cropping and sample balance at the component level.(2)The crack recognition model CrackNet and concrete spalling and rebar leakage recognition models were constructed based on U-Net and VGG-16,and an independent damage pixel-level semantic segmentation database was constructed using RC component test photos as the data source,and a crack width measurement optimization algorithm with CrackNet output as mask and Otsu was proposed.The influence of camera shooting angle on crack width measurement was analyzed,and the reverse perspective error correction was applied to effectively reduce the crack width measurement error caused by the camera’s optical axis and the normal vector of the component plane are not parallel.(3)Three node failure modes associated with RC framework node material,geometric factors,loading mode and apparent damage are introduced,and a node failure pattern database is established by collecting node damage images from relevant literature and manually labeling damage,and by comparing four commonly used classification models,PatternNet based on ResNet is confirmed as the target model for node failure mode classification.By comparing the recognition effect of the model trained on the original image dataset on the node failure pattern,it is verified that the accuracy of component failure pattern recognition can be effectively improved based on component recognition and damage recognition.(4)Through a four-layer RC frame structure shaker test,the crack width measurement effect based on UAV was tested,and the maximum crack width measurement of RC frame node area was realized by applying the crack skeleton line segmentation algorithm through cracks,and the feasibility of RC frame node member identification,damage identification and failure pattern recognition based on UAV was verified.The basic workflow of post-earthquake safety emergency assessment of frame structures based on three identification is proposed.(5)The effect of SIFT feature point matching method,template matching method and phase correlation method in measuring the deformation time history of the structure under the two scenarios of camera rest and movement with the structure was compared by ordinary action camera combined with the target to monitor the shaker response of the five-layer steel frame structure. |
PDF Full Text Request