Study Of Intelligent Construction Site Management Based On UAV And Computer Vision

With the rapid development of engineering construction,the number and scale of engineering projects under construction have become increasingly large nowadays.Scientific construction management is an important part of ensuring the quality,safety and progress of the project.The traditional construction management method is more extensive and the means of management are backward,resulting in high management costs,low efficiency,large safety risks and other issues.Therefore,it is of great significance to introduce digital and intelligent construction site management methods.In order to overcome the drawbacks of traditional management methods,the construction site management method based on UAV and computer vision was proposed,the research was based on UAV technology to collect engineering information in an all-round way,based on oblique photography to achieve a refined threedimensional model construction of the construction site,based on digital image processing technology to achieve virtual simulation of the construction scene,and intelligently identify the safety risks of the construction site through UAV inspection.The intelligent construction management method proposed in this paper will promote the construction of smart construction sites,realize the visualization,informatization and intelligent construction management of construction sites,and have good engineering practical value and social value,and the main contents are as follows:(1)The authors study the refined three-dimensional model reconstruction method of the construction site based on the oblique photography of the UAV,and the method of quickly obtaining the construction site information through non-contact is realized.The multi-view image of the construction site is obtained by the UAV oblique photography mode,and the refined 3D model of the construction site is constructed based on the Structure from Motion(Sf M)and Multi-view Stereo(MVS)matching algorithm,and the accuracy of the reconstructed model in a single direction is better than that of ±3.6 cm.The construction site point cloud is calculated by Cloth Simulation Filter(CSF)algorithm to complete the extraction of engineering information,the amount of earthwork between the ground and the two phases of the design base point cloud model is accurately measured,and a three-dimensional real-life model of multiple consecutive construction key nodes is constructed.(2)The construction site panoramic image stitching method was established,realizing the virtual simulation of the construction site reality based on digital image processing.The image is registered through the aerial photography and attitude information of the UAV,and the attitude data error correction is c arried out by the beam method adjustment method to optimize the accurate registration model of the image,and the Poisson image fusion method of the best suture line is used to eliminate the splicing seam and ghosting,obtain the panoramic image of the construction site from the perspective of the UAV,realize the virtualization of the construction scene and immersive roaming experience,and integrate the construction drawings to help the construction organization plan and arrange,and effectively promote the construction process.(3)The safety risk identification method of construction site is constructed based on deep learning to realize intelligent safety monitoring of intelligent construction site.Aiming at the problems of complex construction site environment and small target picture,the author makes a training set of model suitable for UAV image small target recognition.The model of Faster RegionConvolutional Neural Network(Faster R-CNN)with feature fusion is built.The information of all feature layers is deeply utilized.Safety risks are accurately identified.Identify the wearing status of workers’ safety protection equipment in the construction site,whether the scaffold buckle is effective,and whether the protection of construction openings is in place.The mean Average Precision(m AP)of the model reaches 91.71%,which realizes the accurate identification and classification of safety risks.