Track Defect Detection System Based On UAV Image Acquisition

In order to solve traffic congestion and traffic pollution,large-capacity,efficient,punctual and safe urban rail transit(subway,light rail,tram)has become the ultimate solution for urban transportation development.Manual inspection has a series of major problems.In the subway inspection industry,more portable and small-scale inspection equipment based on technology are needed.Therefore,this paper designs a track defect detection system based on UAV images.The main tasks are as follows:In the preparation stage of the project,this article first inquired about the current four mainstream track defect detection schemes;studied the commonly used subway inspection schemes,went deep into the first line of subway inspections,solved the difficulties and pain points encountered in current inspections,and compared them with several Inspectors with many years of working experience conducted in-depth communication and exchanges.The author extensively analyzed the currently commonly used subway inspection schemes,clarified the service and efficiency standards of the inspection system,and expounded the ideas of this article with several inspectors with many years of service,and initially verified the correctness and feasibility of the inspection system It also designed the system structure of the intelligent drone for subway operation,maintenance and inspection,including: the drone platform and the intelligent detection subsystem.This paper establishes the data center of the inspection system,adopts the micro-service architecture in the software part,draws the software architecture diagram in detail,supports high concurrency,high expansion,and high availability,and explains the module information in the architecture diagram in detail;in line with current Internet companies Mainstream technology stack and development process.This article describes a process and method for the preprocessing of target pictures and target positioning.Taking into account the actual interference factors,adding the prior information of image segmentation and the construction information of subway tracks,the fasteners can be located by positioning the sleepers.Locate the fasteners and rail surfaces separately to minimize the effect of background grayscale on the overall picture.The emergence of convolutional neural networks(CNNs)has accelerated the development of computer vision in many ways.However,most existing CNNs rely heavily on expensive GPUs(graphics processing units).In this article,a compact CNN model is used,which can not only achieve high performance in the detection of small defects,but also run on a low-frequency CPU(central processing unit).The model in this paper consists of a lightweight(LW)bottleneck and a decoder.Using methods such as reduced convolution,lightweight,bilinear interpolation,etc.,compared with the MobileNetV2 model,the accuracy rate is improved by about 5%,and it requires less calculation and shorter time.This paper hopes to add a new track detection technology and method,think about the whole life cycle of Metro maintenance and inspection,and become a tipping point in track detection through differentiation and granularity,so as to endow the influence of traditional Metro maintenance and inspection industry and innovate the ecosystem of Metro maintenance and inspection industry.Based on this cognition and mission,the author hopes to develop a safer and more accurate UAV detection system to strengthen the daily maintenance and inspection system of metro,which can avoid the current pain points and problems,cooperate with the existing detection equipment,restructure the key path of Metro maintenance and inspection,so as to adapt to the ever-changing dimension of Metro detection and strengthen the industrial upgrading.