Research On Surface Crack Detection Technology Of Large Buildings Based On UAV Vision

With the development of society and the growth of the population,the rise of high buildings has been blooming.The exterior walls of the buildings not only play an aesthetic role,but also protect the materials such as steel bars in the walls from being accelerated and rusted.It is necessary to inspect the external wall during the completion inspection and maintenance of the building or after the building.For taller buildings,the use of traditional manual inspection methods for periodic inspection and maintenance of external walls is greatly limited,and there are problems such as low detection efficiency,high cost,low detection accuracy,and large safety hazards.Through research,this paper proposes a large-scale building surface crack detection technology based on drone vision.The research of this subject has great practical application value.The unmanned aerial vehicle combined with visual technology can check and maintain large buildings with flexibility and light weight.It can make up for the shortage of traditional manual detection methods,and has high inspection efficiency and detection accuracy.High and high safety factor.This topic is based on the use of drones combined with visual as a research platform.First,build a large building exterior image system.Then the experiments based on artificial design features combined with BP neural network and deep learning convolutional neural network for wall crack detection are compared and analyzed.Finally,the crack region detected in the large scale image is processed by image,using the NEAR model proposed in this paper.The crack connection algorithm of the external wall surface realizes the crack connection,and the crack after the connection will be more favorable for the classification and quantification of the crack later.The main research results of this paper are:The UAV image acquisition system platfonn was built,and the modules of the UAV image acquisition system were introduced in detail.To create a wall crack database.The wall crack database is constructed by controlling the drone’s flight in the school and photographing the external wall image of the building and downloading the network.The crack target in the image is calibrated as a positive sample of the database,and the non-crack target is calibrated as A negative sample of the database.In the large-scale map crack detection,the method based on artificial design features combined with BP neural network and the method based on deep learning convolutional neural network are compared through experiments.The LBP and HOG algorithms are used to extract the characteristics of the crack’s database and send them to the BP classifier for training and testing.The current mainstream deep learning convolutional neural network models,Faster R-CNN and YOLOv3,are used to train and test the crack’s database respectively.The effect of crack target detection in large scale maps.To compare the effect of two methods on crack target detection in large scale maps.To evaluate crack that have been detected.The image processing of the cracked area has been carried out,and a crack connection algorithm for the exterior wall of NEAR model has been proposed for the crack initiation problem in the crack extraction process.A projection feature based method was used to classify,quantify,and evaluate the degree of damage.