| With the rapid development of science and technology,the trend of intelligence and automation in various industries has swept the world.my country has also increased investment in infrastructure construction,but at present,the degree of automation of domestic road maintenance equipment is still relatively backward.Among them,the road marking and its maintenance work still adopt the semi-automatic operation mode,especially in the damage detection and repair of road marking,it is urgent to improve the modernization level.Therefore,this subject starts the research on damage detection of road markings.In view of the rapid development of deep learning technology based on deep neural network in recent years,especially its image processing method has been widely used in target detection,semantic segmentation and dynamic tracking,therefore,this topic uses the improved YOLOv4 target detection algorithm and selects U-Net Research on damage detection of road markings by network;in addition,a damage detection system for road markings is designed by using the damage detection algorithm proposed in this paper.For road marking detection.Since the network structure of YOLOv4 is relatively complex,which is not conducive to the deployment of the edge of the network,this paper first studies the lightweight of the entire network.Using a deep separable network greatly reduces the amount of parameters in the network.However,after a lightweight and improved algorithm,the detection results of multi-category markings will be unbalanced.After incorporating the attention mechanism into the network model,it is greatly improved.The balance of multi-class detection is improved.In order to further enhance the performance of the model,the K-Means++ algorithm is used to select the prior frame to obtain a more suitable prior frame;the loss function is optimized to accelerate the convergence speed of the model and enhance the learning ability of the model.In order to verify the performance of the improved model,the road markings are detected using the existing target detection algorithm and the improved target detection algorithm.Comparing the experimental results,it is found that the improved target detection algorithm has good performance and can detect road markings in real time and accurately.For the detection of damaged road markings.This paper uses the existing U-Net semantic segmentation algorithm to segment the road markings semantically,and obtains the semantic segmentation model of the road markings.In order to further detect the damage of road markings,this subject first uses the proposed target detection method to detect the markings,and obtains the type,area and accuracy of the markings;then carries out semantic segmentation of road markings,and uses the obtained Road marking positioning information;finally,using the category information and location information of road markings,combined with the positioning information of semantic segmentation,the damage detection of road markings is carried out,so as to obtain the damaged area of road markings and the damage rate of road markings.Finally,according to the road marking damage detection algorithm,a road marking damage detection system is designed,which can automatically detect road marking damage according to the input road scene information. |
