Vision Inspection System Of The Urban Sewer Defects Based On The Mobile Robot

With the rapid expansion of cities,it is more and more important to ensure the safety of urban health and living environment.Among them,one of the important factors affecting the life of urban people is the smooth discharge of urban sewers.Then,due to the factors such as the lack of repair of urban sewers,the increase of emissions,the construction of infrastructure and geological movements,the urban sewers appear different degrees of damage or defects.If it can not be found and repaired in time,it will cause urban disaster risks.At present,the defect detection of urban sewers is mainly carried out by the maintenance personnel entering the urban sewer inspection and repair,the CCTV video assisted by the mobile inspection robot and its manual analysis to interpret and discover the damage condition of the urban sewer wall.These methods have the problems of high labor intensity and endangering the health and safety of personnel.Therefore,this paper focuses on the key technologies such as the visual intelligent detection system scheme of urban sewer defects based on inspection robot,the visual intelligent detection algorithm and system of urban sewer defects.The main research work and results are summarized as follows :(1)Aiming at the problems of low detection efficiency and high labor cost in manual interpretation of urban sewer defect inspection,an urban sewer defect detection system was built.According to the requirements and functional analysis of the urban sewer inspection system,the overall scheme of the system is determined,including system composition,hardware and software structure.Finally,based on the analysis of the dark environment of urban sewers and the complex road conditions,the image acquisition system is analyzed and selected.(2)Aiming at the problem that the existing algorithms cannot balance high precision and small volume,a visual detection algorithm for urban sewer defects is designed.Using deep separable convolution instead of ordinary convolution greatly reduces the parameter quantity and calculation amount of the model,and controls the network depth and channel number to reduce the complexity.The final model has a depth of 22 layers,a parameter number of only780,000,a memory footprint of only 10.5 M,a floating-point operation of only 68.45 million,and a detection speed of only 0.126 s.It is superior to most existing models in terms of lightweight indicators and achieves the goal of small model volume.At the same time,in order to achieve high-precision goals,a residual structure with better convergence and expression ability is used as the basic architecture.Three residual modules of the receptive field are designed to avoid repeated expression of information at the same scale.On this basis,three short-path cross-scale fusion modules are designed.The feature reuse and residual module are combined,and the local information is obtained by image splitting and then directly fused with the module output layer.Finally,aiming at the problem of data imbalance,the data set is divided into two parts,and the two-stage method is used for detection.The experimental results show that the anomaly detection accuracy of the first stage can reach 95.1 %,the multi-classification defect detection accuracy of the second stage can reach 89.7 %,the recall rate can reach 88.7 %,and the F1-score can reach 89.2 %,which is better than the existing classical CNN network.(3)In order to improve the integration of urban sewer detection system,the development of visual system software for urban sewer defects was completed.Py Qt5 is used as the software design platform.The software system integrates the hardware system and defect detection algorithm proposed in this project.After the user logs in,the operating parameters of the inspection robot can be set and adjusted.After the defect is detected,the detection results of the urban sewer are output online.