Research On Detection Methods Of Road And Bridge's Diseases Based On UAV Imagery

Road and bridge diseases are one of the main basis for arranging inspection and maintenance work.The number and severity of road and bridge diseases directly reflect the road and bridge health status.Moreover,changes in road and bridge disease indicate the surface or deep lesions,thereby affecting the road and bridge life and safety factors.Linear Crack disease,Reticular Crack disease and Hole disease as early diseases of road and bridge,are common disease on the surface,and their parameters(such as the shape,area and location of the Linear Cracks,etc.)are important parameters for assessing the health status of roads and bridges.Traditional road and bridge inspection is a combination of road and bridge inspection vehicles and inspectors' visual inspection,Unmanned Aerial Vehicles(UAV)as a new sensor platform have obvious advantages,such as easy operation,flexible viewing angle,low cost characteristics.With the development of low altitude photogrammetry and image processing technology,the detection method of road and bridge disease has undergone significant changes,and the changes of detection platform and image processing methods are bound to have a certain influence on the whole road and bridge disease detection system.Therefore,a comprehensive analysis of UAV-based road and bridge disease detection methods and their accuracy assessment,is of great significance to the assessment of road health and the corresponding inspection program arrangements.This research is based on the data of road and bridge obtained by hand-held shooting and the image data captured by SONY ILCE-7R camera equipped with UAV DJI-900.Firstly,we use UAV image processing to eliminate the ambiguity of UAV images,and then use the hand-held camera to obtain the road and bridge disease image training Linear Cracks,Reticular Cracks and Hole these three types of multi-component deformation model of early and basic road and bridge disease.The multi-component deformation model is used to detect the road and bridge disease area on the UAV images,and based on the disease area to extract the road and bridge disease information by edge detection methods.Finally,the development of road and bridge disease detection system based on UAV images is introduced and verified.By analysing the methods of UAV image processing,road and bridge disease model,model detection and road and bridge information extraction,and road and bridge disease detection system,drawing the following conclusions:(1)An innovative detection technology of road and bridge diseases based on UAV images is proposed.Aiming at the characteristics of UAV images,this paper compares various image processing algorithms to improve the image quality.The multicomponent deformation model,which can describe the morphological,structural and texture characteristics of road and bridge disease,is verified its simulation of the feasibility of road and bridge disease,and road and bridge disease model library is established.Then the road and bridge disease model is used to detect the image area of road and bridge disease on the UAV image,and the edge detection algorithm is used to extract the road and bridge disease information in the image area.Methods proposed aboved greatly improve the effect of road and bridge disease detection and information extraction.Finally,the function,algorithm and implementation verification of road and bridge disease detection system are expounded.(2)Through the study of UAV image characteristics and image preprocessing methods,UAV images quality are obviously improved.By comparing the fuzzy probability model and the standard anisotropic diffusion under the condition of multiparameter condition,it is proved that the standard anisotropic diffusion function performs better in eliminating the ambiguity of UAV images while preserving the edge characteristics of the image.(3)Through studying the image characteristics and feature expression of road and bridge disease,a multi-component deformation model,which simulates the image shape,structure and texture characteristics of road and bridge disease,is proposed to apply in road and bridge disease detection,and generalization performance of disease model is compared.The results show that the Linear Cracks model has a good generalization ability and the average accuracy of the model is AP = 0.64891.The simulation function of the Reticular Cracks model is poor and the average accuracy is AP = 0.5658.The Hole model has the strongest generalization ability,average accuracy AP = 0.68254.In order to extract the road and bridge disease as far as possible,if the roadway disease detection rate is required higher,when the recall value is 0.6,the precision is between 0.2 and 0.4.(4)By applying the disease models to the UAV image,there is a difference between the setting of the response threshold and the detection effect,and the edge segmentation effect is also significantly different.The results show that the response of Linear Cracks is the highest when the response rate is-0.95,the highest rate of Reticular Cracks appears at the response of-0.99,the Hole Cracks response is-0.9.Co mpared with the classical canny algorithm and sobel algorithm,the idea proposed by this paper can extracted the edge and shape information of the road and bridge disease more accurately by using the local adaptive threshold segmentation and the minimum connected domain removal method.(5)Through the programming of road and bridge disease detection system based on UAV images,system design,function realization and instance verification are elaborated.The results show that the algorithm and methods proposed above are effective and reliable,and the system has good function and good application prospect.