Research And Implementation Of Road Crack Recognition Algorithm Under Unfavorable Vision

With the rapid development of society,the relationship between cities is getting closer and closer,and the essential factor is the perfect transportation system in China.The key object of daily maintenance of transportation system is highway system.Many roads have been used for a long time,and their potential safety hazards gradually appear.Among the hidden dangers of roads,cracks are one of the defects that highway systems need to maintain frequently.Accurate identification of road cracks in actual complex environment provides great convenience for highway maintenance and management.At present,the detection algorithms used in engineering applications are mostly based on image processing,which can meet the detection requirements in common environment.In practical work,there are various unfavorable visual environments due to weather or illumination factors,resulting in the low quality of the collected images and affecting the detection results.In this paper,an algorithm of road crack recognition based on unfavorable visual conditions is designed,which combines image enhancement technology with target detection algorithm for two types of images in foggy days and at night.The following describes some of the main work:(1)Make fracture data set under unfavorable vision and deal with it accordingly.In the process of road maintenance,all kinds of cracks on the road surface are the recognition targets.The road crack sample data set used in this paper comes from RDD-2019 data set in the challenge of road damage detection and classification in 2019,and contains road images with various backgrounds.The sample data is expanded by rotation to improve the generalization ability of the model.(2)According to the characteristics of adverse visual environment such as foggy days and low illumination studied in this paper,it is very important to select appropriate image processing methods for adverse visual image enhancement.In this paper,several popular defogging algorithms and low illumination enhancement algorithms are compared,and finally SSR algorithm with relatively excellent subjective feelings and objective evaluation criteria is selected.(3)Select the target detection algorithm suitable for crack details and combine it with image enhancement algorithm.After analyzing the principle of each algorithm and comparing the performance of the models,Faster R-CNN model is adopted as the network model of crack detection.The detection effects of normal environment,unfavorable visual conditions and enhanced crack images are compared.