Bridge Appearance Construction Quality Inspection Based On Deep Learning And Depth Camera

With the development of the times,railway bridges play an important role in economic development.Its construction and maintenance are very important.At present,most of the bridge quality detection relies on the construction personnel,construction supervision,operation and maintenance personnel’s eye detection or sensor detection,which is time-consuming and laborious with low accuracy.In this paper realize to detect bridge cracks,reinforced nudity,surface spalling,corrosion and weathering besides measuring the geometry of the bridge to determine whether can produce safe hidden trouble.And building a prototype software system is connect to the depth camera which can collect the data,store the data,display the data,query the data and so on,it provides favorable support for the inspection and maintenance of bridges.The main work of this paper are as follows:1.Achieving the detection of bridge defects.First,the bridge cracks,reinforced nudity,surface spalling,corrosion and weathering are made into standard data sets.Second,using data enhancement and data cleansing and adding certain types of data to complete data sets.Third,YOLOv5 is used as the basic network model of target detection,then the data sets are used to train YOLOv5 which achieve the 81.4% of the m AP.Fourth,it proposed to add channel attention mechanism in order to improve the YOLOv5 model,By comparing the results of improved YOLOv5 model with the YOLOv5,the m AP value is improved by 4.2%.So the detection result is better.2.Realizing to measure the geometry of the bridge.This method is non-contact detection which the depth camera is used to achieve and use the introduction of target to get the original design value.the depth camera can obtain 3D information,these information can use image processing technology to obtain measured values,comparing the design value with the measured value can get the conclusion that relative error less than 2%,compared with the measurement method based on two-dimensional image,the average relative error is reduced by 2.25%.And it is more convenient and fast than using professional measuring equipment such as sensors.3.Building the prototype software system of bridge construction quality inspection.The bridge defects identification algorithm and bridge geometry measurement algorithm are integrated,and the external depth camera is connected to this system which can realize the functions of data collection,visual display of data results,data statistics,data storage and data query.The methods of bridge quality inspection implemented in this paper have achieved better results under the existing experimental conditions which make the bridge quality inspection more accurate,convenient and fast.In the prototype system designed at the end of this paper,the component design idea is adopted,which can easily add and delete functional modules and carry out later maintenance.It is hoped that such a system can be used in the detection and maintenance of bridge quality in the future,or combined with UAV and robot can be used in the place where the environment is bad or endangers the personal safety of inspectors which make a little contribution to the cause of bridge quality inspection at this stage.