Research On Road Crack Type Recognition Method Based On Drone Image

In the past 10 years,highway construction has developed rapidly,and mileage has been increasing.But the investment growth in highway construction has stabilized,and the maintenance of existing highways has become a top priority.Road cracks are the earliest manifestation of highway diseases.Early detection of cracks and timely repair can prevent further expansion of the disease,thereby reducing capital consumption and reducing the cost of highway maintenance.At present,the detection of highway cracks includes two methods: manual detection and automatic detection of vehicle equipment.Manual detection is the most traditional.Everything in manual detection is mainly artificial,which consumes a lot of human resources.The automatic detection of on-board equipment generally requires firstly sealing the road to be tested,and then completing its detection work.The detection efficiency is high,but the cost is high and there are certain security risks.In recent years,UAV technology has been applied to many fields such as reconnaissance,agriculture,forestry,and disaster rescue due to its advantages such as fast,accurate,and safe.Therefore,this article will extend the application of drone technology in road disease detection.This paper proposes an efficient and accurate road crack detection method for complex UAV image data.The main research work of this article is as follows:(1)Method for acquiring road image data.In this paper,a drone is used as a data acquisition device to obtain road images quickly and in real time.Considering the flight safety of the UAV and the terrain limitation of the research area,its flying altitude is generally high,and the types of ground objects contained in the acquired images are more complicated.(2)Road image data processing method.The types of ground objects included in the drone image are complicated.And it is necessary to eliminate the interference of factors such as vehicles,trees,and building construction.Image preprocessing includes three steps:graying,image segmentation and filtering and denoising.Graying can reduce the drone image data occupying memory,which can promote the processing efficiency of subsequent operations such as crack segmentation and classification.In order to separate crack elements from drone images,image segmentation has become an important step.This paper compares the effects of traditional edge detection operators and the four gray prediction models on crack segmentation.Finally,it is found that the traditional image segmentation effect is average,and the Verhulst model in the gray prediction model is better than the traditional segmentation method.The segmented image has a lot of noise.Filtering and denoising can eliminate the influence of noise.After comparing the denoising principles and effects of mean filtering,median filtering and adaptive Wiener filtering,the adaptive Wiener filtering with better denoising effect and more applications is used for noise removal.(3)Road crack type identification methodThe processed UAV images are used to identify and extract road crack information based on road crack image characteristics.Then,follow-up operations are performed on the extracted information,mainly to connect fractures and cracks and remove mis-extracted elements.Through comparison and analysis of closed operation,KD tree and minimum spanning tree method,Prim’s minimum spanning tree method is finally used to achieve the connection of fractures and cracks.According to the characteristics of the element’s connected domain,the area and rectangularity are used to remove the point and area mis-extracted features.When identifying the types of road cracks,this paper selects feature values not only traditional geometric features,projection features,and distribution density features,but also direction features.In this paper,a section of city road in Shijiazhuang City,Hebei Province and a village road in Baoding City,Hebei Province are used as research areas to experimentally verify the detection method of this article.After experimental verification,the traditional feature extraction method for road crack type recognition accuracy is 84.92%,and the method proposed in this paper reaches 90.93%.Experimental analysis confirms the practicability,accuracy and effectiveness of the intelligent detection method in this paper.This shows that the directional feature is feasible and necessary as one of the features of the crack type recognition classifier,with high detection accuracy and strong practicability.