Pavement Crack Detection based on yolo v3

Abstract

In recent years, with the rapid development of the economy, road construction has entered the stage of coexistence of construction and conservation. Even road maintenance has become a major aspect of road construction. The government invests a lot of money in road maintenance every year. Therefore, it’s very important to detect cracks on the road surface in order to reduce the cost of maintenance. Aiming at the problems of poor real-time performance and low accuracy of traditional pavement crack detection, and using the advantages of deep learning network in target detection, a method based on yolo v3 for pavement crack detection is designed. The collected pictures are manually marked, and the network model is obtained through yolo v3 network training. Finally, the cracks are detected and verified by the obtained model. The accuracy of crack detection in this work reached 88%, and the crack detection speed was also improved compared with the traditional identification method.