Pavement Crack Detection Of Aerial Image

Pavement crack detection algorithm in the highway quality inspection and evaluation,with accurate and efficient advantages,quickly get a wide range of applications.However,the existing application scenarios of the algorithm are images collected by the on-board method,with the efficient and convenient unmanned aerial vehicle acquisition,it is no longer suitable for the task of crack recognition in aerial images.Therefore,this thesis focuses on the pavement crack detection algorithm for aerial images.Based on the parameter of UAV platform,this thesis analyzes the correspondence between the camera pixels and the crack target,selects suitable HD camera acquisition device.The aerial images of the road surface are acquired through many experiments and made for deep learning training image data sets.And crack detection software design and offline user interface are also displayed.In this thesis,aiming at the problems of interference and noise in image recognition of aerial pavement,a pavement crack detection algorithm based on image segmentation is put forward.According to the difference of gray level distribution of the surface area and the roadside landscape area,a method of regional growth based on multi directional fitting and threshold segmentation in HSV color space for road region segmentation is proposed.The single channel pavement which contains integral crack information is extracted,the large area of interference is eliminated by the sliding window filtering,and an edge detection algorithm based on saliency analysis to recognize the crack fragment of pavement is proposed,realizing the distinction between complex cracks and pavement texture noise.A single pixel skeleton is extracted for the suspicious area of the crack,and the length of the skeleton is calculated.The experimental results show that the proposed method can effectively remove the interference and noise in the image,and well identify asphalt pavement cracks.The accuracy of the screening of suspicious fractures is above 80%.The classification accuracy is over 85%.The accuracy of length measurement is over 75%.In this thesis,in order to improve the robustness of the traditional image processing algorithms,a pavement crack detection algorithm based on deep learning is put forward.A rough road segmentation layer based on K-means algorithm is proposed,which has the characteristics of high recall.By analyzing the mature network structure and combining with the detection of small targets in complex scenes,a feature extraction network structure is designed.Referring to the region proposal network of Faster-RCNN,this thesis designs a region proposal network that satisfies the feature extraction network.The classification network is used to locate and classify the preliminary detection results,and the single pixel skeleton of the crack is evaluated and analyzed in the detection result area.The experimental results show that the proposed algorithm is robust against complex scenes and the accuracy of crack detection is about 89%.