The Research On Exterior Wall Crack Detection Technology Based On Unmanned Aerial Vehicle And Improved Mask R-CNN

Building exterior wall cracks detection is an important part in engineering construction,but the existing exterior wall cracks detection technologies are still based on manual detection,there are low efficiency,high cost and low security shortcomings.With the increase of the height and width of the building,the detection range increases,and it is difficult to complete the detection task only by manual detection and contact measuring instruments.Therefore,the design of an intelligent exterior wall cracks detection method is an inevitable requirement for the development of Intelligent detection of building exterior wall,and has important innovative significance for the research of new building quality detection technology.Based on the existing theoretical results,this paper has carried out the following specific work:(1)The plane image features of transverse cracks,vertical cracks,oblique cracks and irregular cracks were analyzed by collecting and studying a large number of cracks images of external walls of buildings by UAV.The possible difficulties in cracks detection and the improvement direction of neural network were elaborated.(2)Five object detection networks,Single Shot MultiBox Detector,You Only Look Once v3,You Only Look Once v4,Fast R-CNN and Mask R-CNN,were trained,and the cracks detection effects of those were compared.The results shown that Mask R-CNN with instance segmentation function has the best effect on cracks detection.To solve the problem that the low utilization rate of Mask R-CNN for the characteristics of the bottom layer of cracks,DenseNet was introduced to replace the backbone network of Mask R-CNN.The improved Mask R-CNN greatly improves the cracks detection effect,such as the accuracy was improving and the missed detection rate was reducing.(3)A cracks detection method of exterior wall by UAV was proposed.In the aspect of exterior wall image collection,the UAV route planning scheme was designed,and the video recording method was proposed to replace UAV hovering photography for image collection,which improves the efficiency of exterior wall cracks detection.In the aspect of image processing,an image extraction scheme is designed according to the requirements of UAV visual field height,flight speed and image repetition range.The cracks of the external wall were detected and marked with the improved Mask R-CNN.In the aspect of exterior wall modeling and cracks detection data storage,the advantages and disadvantages of 3D modeling and plane modeling in cracks data preservation are compared,and the image stitching technology was used for plane modeling and storage of exterior wall crack labelings.(4)The crack detection technology of UAV was used to detect some walls of Changsha University of Technology Library.The detection results shown that the method has the advantages of low cost,high detection efficiency and high security,and can effectively detect the cracks of exterior wall images.The exterior wall plane modeling effect is good,through the regular detection and modeling of the exterior wall,the development of the settlement cracks,contraction cracks and other exterior wall cracks can be timely comparative analysed,predicted and prevented,which has practical application value.