Research On Intelligent Road Crack Detection System Based On Computer Vision

With the rapid development of road construction in China,road inspection and maintenance work is becoming increasingly heavy.Crack detection is an important link in road maintenance work.In the early stage of cracks,timely and effective detection and maintenance measures can avoid road structure damage and traffic accidents.At present,the traditional road crack detection work is mainly based on manual detection methods,which are inefficient,time-consuming and have low detection accuracy,which is difficult to meet the needs of current road development.In recent years,deep learning technology has achieved remarkable results in the field of computer vision,and has been widely used in transportation and industrial production scenarios.The deployment and development of deep learning platforms based on embedded devices,smart phones,FPGAs,clouds,etc.are also gradually developing.It provides the possibility for an intelligent road crack detection system based on deep learning technology.In view of the shortcomings of the traditional road crack detection system,the main research work of this paper is as follows:First of all,in order to adapt to the application requirements in multiple scenarios,this paper designs and implements an intelligent road crack detection system based on deep learning technology in combination with the current related technologies and deployment solutions in the field of deep learning.The system mainly includes two deployment solutions based on cloud servers and micro PC terminal devices.The cloud-based crack detection solution uses B/S mode to realize the interaction between the client browser and the server.Users can rely on the server to perform deep learning through the browser Task,using cloud server computing performance to achieve online training and crack real-time detection of YOLOv3 detection algorithm.In order to solve the problem that the deep learning algorithm is difficult to deploy in terminals with limited power consumption,the solution based on the micro PC terminal uses a lightweight network Tiny-Darknet to build a miniaturized model Tiny-YOLOv3 for crack detection,and model optimization and inference through the Open VINO platform accelerate.Secondly,this article collects road crack images to build a project data set,which is mainly divided into three stages:image preprocessing,data annotation and data screening.On the project data set,the performance comparison of the YOLOv3 and Tiny-YOLOv3 algorithms under the two deployment schemes was respectively performed,and the superiority of the two algorithms in terms of detection efficiency and accuracy compared to other detection algorithms was verified,and the cloud and terminal equipment were given.The inference detection effect of the deployment plan.In addition,this article is based on the performance comparison experiment of the micro-PC terminal equipment under the OpenVINO platform.Compared with the TensorFlow framework,the Tiny-YOLOv3 algorithm infers the speed by 11.2 FPS on the OpenVINO platform.Finally,in order to verify the feasibility and effectiveness of this system for road crack detection,this paper tests and analyzes the comparison effect of the two scenarios based on cloud and terminal equipment deployed in actual scenarios.High,but the micro PC terminal solution is easy to deploy,easy to implement,has good real-time performance,and can achieve a detection speed of 17.6 FPS in actual scenarios.