Pipeline Defect Based On RGBD Video Image Automatic Detection And Recognition

As a common means of transportation of gas and liquid,pipeline is prone to block,crack and other defects due to its long service life and aging,which leads to various safety accidents.Therefore,a lot of labor cost is needed to carry out pipeline inspection and maintenance.However,the detection of pipeline defects is limited by the environment.The traditional manual detection has high difficulty coefficient and low detection efficiency.From the perspective of computer vision,this paper explores the technology of automatic detection and recognition of pipeline defects,and proposes a method of automatic detection and recognition of video defects based on deep neural network,so as to reduce the labor cost.In general,the pipeline inspection robot will integrate many kinds of sensors,load intelligent mobile carriers,or carry operation devices and nondestructive testing technology.With its small design,it can enter the complex non structural pipeline environment beyond human’s reach,collect the images in the pipeline through the camera,carry out image processing,analysis,defect identification,mark alarm and so on in real time 。Using the pipeline robot to detect and maintain the pipeline can accurately analyze and judge the corrosion,obstacles,fractures,floating objects and other defects of the inner wall of the pipeline,and can accurately locate the pipeline defects using the positioning system of the pipeline robot.This paper summarizes the research of pipeline detection technology,depth network technology,image processing technology and machine learning field at present,analyzes the theoretical rationality,effective operation and practicability of this research,and clarifies the significance of automatic identification of video pipeline defects based on RGBD.Secondly,this paper analyzes the noise models that affect the image quality,such as Gaussian noise,salt and pepper noise,and focuses on the image denoising methods,such as mean filter,Gaussian filter,median filter,etc.through experimental means,it is clear that Gaussian filter is an effective means of dealing with Gaussian noise,median filter is an effective means of dealing with salt and pepper noise.Thirdly,by constructing neural network,this paper compares the main models of convolution neural network.Through the comparative analysis of loss rate,it can be concludedthat Alex Net model or the classification effect of pipeline defect pictures is better.After expanding the sample set,the accuracy of model training is improved by nearly5%,and the accuracy of model verification is improved by nearly 7%.Fourth,this paper analyzes two methods of image segmentation,and verifies that adaptive threshold segmentation has better advantages for image processing.Finally,this paper proposes a video analysis method combining video difference and adaptive moving rate,which is helpful to the rate of defect recognition.In this paper,the research on automatic detection and recognition of Pipeline Defects Based on RGBD video image is proposed.The use of image intelligent processing technology not only reduces the subjective factors,but also improves the detection efficiency of pipeline defects.At the same time,intelligent detection will also be an indispensable part of building a smart city in the future.