Algorithm Research On Pipeline Defect Inspection Based On Machine Vision

Draining pipes works like human vessel, but similarly it will break down once the pipe or vessel gets blocked. So it is an essential part to repair and to inspect the pipes. Currently, CCTV(Close Circuit Television) inspection technique is one of the most commonly used ways to inspect underground pipes. But this method requires human operation whether in grabbing or discerning the image, as a result, human visual asthenopia and distraction often affect the accuracy and precision needed in pipe inspection. It does not meet the intelligence and involves much human subjectivity. The paper mainly researches on automatic image grabbing and automatic interpretation based on machine vision by concerning the drawbacks in CCTV pipe inspection method. So automation level and efficiency will correspondingly enhanced, together with human labor work reduced.Research inspecting defects on underground pipes has the following new contents and novelty concerning the characteristics of underground pipes:1.Apply an improved low-pass filtering difference method to extract functional defects on pipes by considering the complex circumstances in capturing image using CCTV method. The algorithm can avoid the influence imposed to image under complex background and asymmetrical lighting status. The method has a desired segmentation result towards functional defects and is much superior to other traditional segmentation methods.2. Analyze the distinct regions on defects by choosing roundness, compactness, convex-concave degree, angle second order matrix as features to characterize information on pipes inwall and to provide the basis for pattern recognition. Eigenvectors obtained by statistical analysis enable each of them to distinguish different kinds of defects.3. Provide an improved method towards the traditional BP neural network algorithm when performing pattern recognition of defect types. It has applied momentum-adaptive learning rate adjustment, effectively reducing fluctuation in BP neural network learning and improving astringency. Meanwhile, while doing network training on-line learning training method is adopted to avoid being caught into local minimum.4. The novelty is that the study has proposed a new drainage pipe inspection method combing the three-dimensional laser scanning technique with CCTV inspection technology. The proposed method can quickly extract the structural defects on pipes and in the same way compensate deficiency imposed by CCTV inspection technique. Furthermore, it can locate defects on drainage pipes with perceptual intuition and make reinforcement of pipes more concrete.The novelty is that the paper has used low-pass filtering difference method to extract the functional defects on complex background image captured under natural lighting using CCTV. Then the pipe contour image is acquired by three-dimensional laser scanning technique and it uses polar plane comparison method to obtain the structural defects.