Computer-Vision-Based Identification Methods Of Structural Displacement And Cracks

Sensing technology is one of the main problems in structural health monitoring,in which displacement is an important variable of structural health.The traditional contact displacement sensing technology has sparse measuring points and its installation is timeconsuming.With the development of artificial intelligence technology,the application of camera and its robot and UAV will be more common,and the image and video data will be more easier to obtain.The displacement monitoring and recognition method based on computer vision can easily obtain the structural dense displacement data of high spatial density,which is a kind of sensing method with broad application prospects.In addition to displacement,structural surface damage can be identified by computer vision based on image and video data.The research content of this paper is as follows:A single point displacement recognition method based on support vector filter and feature point recognition algorithm is proposed,and the rough position of the target is identified by the support vector filter method,which improves the robustness of the algorithm.The feature point recognition algorithm is studied to identify the exact position of the target and improve the accuracy of the algorithm.The robustness of the proposed method is verified by a shaking table tower model experiment.A structural displacement recognition algorithm based on structural component segmentation and pose parameters is proposed.A recognition network architecture is designed to construct a deformable structural 3D model first and then further identify structural displacement.Two subnetworks are proposed: structural component semantic segmentation subnetwork and structural pose parameter recognition subnetwork.The original video is the input,and the semantic segmentation mask is identified through the feature extraction ability of convolutional neural network.The multi-channel convolution structure provided by Inception V3 network is used to output the structural pose parameters,which are applied to the structural mesh model to obtain the structural displacement results.Structure random deformation image generation algorithm is proposed,using the random structural control node interpolation algorithm for image deformation with labels,established from a few labeled data generated to meet the deep learning training requirements in the number of sample data generation model,through multiple shaking table model test displacement accuracy of the proposed algorithm,The displacement of Tacoma bridge is identified based on video data.A structure displacement identification algorithm based on deformable structure mesh model and control points is proposed.Two subnetworks,control point estimation subnetwork and pose parameter identification subnetwork based on control points,are proposed.The original video is used as the input,and the control point coordinates in the subnetwork are estimated by dense optical flow,and the two-dimensional position of the control point is output.The feature extraction ability of multi-layer perceptron is used to identify the pose parameters of the structure from the coordinates of the control points,and the displacement of the structure is calculated.A three-dimensional model is used as the displacement carrier to ensure the density of the displacement identification.A shaking table experiment with a four-layer frame is used to verify the accuracy of the displacement identification and the robustness of the results against light changes.An algorithm for identifying structural cracks based on self-attention-self-adaptive neurons was proposed,and a self-attention-self-adaptiv neuron was proposed.A gate function was used to transfer the computation of the neuron backward to the part with large value,and a self-attention mechanism was established.Use multilayer perceptron model embedded in the neurons increased since the adaptive mechanism,enhances the identification of individual neurons and the overall network effect,improve surface cracks in a structure based on image recognition accuracy increase fracture identification network for feature extraction of single neurons in effect,the accuracy of the algorithm is verified by the steel box girder crack data.