Research On Bridge Crack Detection Algorithm Based On Image Measurement

As a key facility of transportation,bridges play a vital role in people’s daily travel.However,with the continuous increase of the total amount of highway and bridge infrastructure in China,cracks and other diseases will occur on the bridge surface due to truck overload,traffic accidents,aging of building materials and other factors.Many in-service bridges continue to deteriorate and damage,resulting in bridge collapse accidents.Its condition needs to be monitored and evaluated regularly.The traditional crack detection has the problems of great difficulty,low efficiency and high size measurement error due to the limitation of terrain.Moreover,the previous crack shape segmentation methods are under the ideal background without the interference of obstacles such as fallen leaves,water stains and lane lines,which do not meet the natural attributes in the actual bridge image and are difficult to meet the needs of practical engineering.To solve the above problems,this thesis proposes a bridge crack segmentation and measurement technology based on solov2 segmentation model.Firstly,this thesis selects unmanned aircraft as auxiliary equipment,uses a mountable camera and a three-point laser instrument to measure the crack,so as to obtain the bridge crack image with object distance parameters.Taking UAV as the working platform for detection can overcome the impact caused by terrain and do not occupy the lane.It can adapt to the harsh natural environment,reduce the work intensity and reduce the operation risk.Secondly,this thesis constructs the intelligent extraction model of crack shape through the instance segmentation model based on solov2,which can realize the automatic extraction of crack shape under complex background,and this thesis optimizes the original model to improve the segmentation effect.The specific approach is to introduce the attention mechanism into the original instance segmentation network model solov2,improve the feature extraction ability of the backbone network,and solve the problem of loss caused by different features accounting for different recognition results,so as to obtain more complete crack information,and the segmentation effect of crack shape is more ideal.In the case of single bridge crack shape,In this thesis,the example segmentation of solov2,which introduces the attention module,is calculated on the public data set and the UAV self collected data set.The morphological segmentation and extraction can achieve an accuracy of 87.3%,which is more than 2.1% higher than the baseline method.Moreover,in this thesis,when measuring the size of bridge cracks,with the help of external sensing equipment and image measurement method,this algorithm obtains the crack picture and distance information through the UAV mounted camera and laser rangefinder.Firstly,the crack skeleton is extracted through the central axis calculation method to obtain the pixel size such as crack width that meets the calculation accuracy of bridge crack width.Then,considering that cracks are mostly distributed on the plane,the camera imaging plane may not be parallel to the plane of the crack area,and there is a certain deflection angle,so the text uses the three-point method to measure the object distance.Only by measuring the three-point object distance not on the same straight line,the relative angle relationship between the crack plane and the camera imaging plane can be obtained,so that the image pixel resolution can be corrected,Finally,the accurate calculation of the actual physical size of the crack can be realized.The experimental results show that the data calculated by the method of crack size measurement based on image measurement proposed in this thesis is less different from the data detected by manual crack measuring instrument,the error is within the acceptable range,the measurement accuracy can reach 98%,which can meet the needs of practical engineering and has a good application prospect.