Research And Implementation Of Pavement Damage Detection System For Urban Roads

With the rapid development of the economy,the number of cars in China has increased significantly,and the intelligent driving of automobiles has become a hot research field at present,and road damage detection is an extremely important part of the research field of intelligent automobile driving.The research on road damage detection has been developed for a long time at home and abroad.The early detection method is to detect by embedding a large number of sensors in the car.With the development of computer technology,image processing technology has also been applied to the research of road damage detection.In recent years,deep learning methods have also been gradually applied in the research of pavement damage detection.This thesis mainly takes the detection of pavement damage as the research content,discusses the adaptive improvement method based on the yolov5 target detection algorithm and the method of analyzing the characteristics of the pavement damage part based on the image processing technology,and describes the design and implementation of the pavement damage detection system.The main research content of this thesis includes an adaptive improvement plan for the defects existing in the application of yolov5 algorithm to road damage detection.The improvement content includes three aspects: network structure improvement,loss function improvement and activation function improvement.Difference structure and ghost convolution,cut out focus structure and prediction feature layer branch,improve feature fusion structure and SPP structure,add the improved prediction target frame length and width loss factor in the prediction target frame location loss part of the loss function,and activate the function Improved to the smoothed FRe LU activation function;The training method of the network model and the non-maximum suppression method at the output end of the network model are improved.In the process of model training,based on the mosaic image enhancement method,it is proposed to randomly transform the brightness,contrast and saturation of the images in the training data set.The preprocessing method of randomly adding noise,for the non-maximum suppression method,proposes to use the DIOU method to calculate the overlap degree of the target frame,and at the same time integrates the soft-NMS method to gradually reduce the confidence of the target frame;based on the improved The target detection method is to perform image processing on the detected image of the damaged part of the road,analyze and estimate the area,height or depth of the damaged part,and at the same time realize the estimation of the distance between the damaged part of the road and the car;the road damage detection is described.The realization of the system is mainly based on the improved target detection algorithm.The functions of the system include real-time detection and reporting of road damage,and storage of the location information and damage degree information of road damage.This thesis trains and validates the improved algorithm,and tests the system.The experimental results show that the improved algorithm and the urban road pavement damage detection system designed and implemented in this thesis have good results and have good stability in actual detection.