Research And Implementation Of Pipeline Surface Damage Video Detection Based On Haar-like Feature Fast Extraction

The urban underground integrated pipe gallery is in the stage of rapid development.Due to the influence of transportation objects and uncontrollable factors in nature,the surface of the pipe will inevitably be damaged.Common damages include cracks and holes.In the traditional maintenance method of the pipe gallery,the manual inspection workload is large and the detection period is long;because of the monitoring dead angle in the fixed node monitoring,the pipeline surface scene cannot be covered,and the monitoring cost is too high;the magnetic flux leakage detection will be affected by other cables in the pipe corridor.Efficient and accurate detection of pipe surface damage in pipe corridors is an important issue to be solved.Aiming at the deficiencies of the above detection methods,this thesis proposes a damage detection algorithm based on Haar-like feature fast extraction,which realizes the video detection of pipeline surface damage.Main works of this thesis are as follows:1.For the narrow space of the pipe gallery and the insufficient lighting.This thesis adopts the schme of the indoor autonomous mobile robot equipped with a 720 P camera for video capture.By extracting the I frame in H.264 video coding,the image acquisition efficiency is improved,and the real-time requirement of video image detection is guaranteed.At the same time,according to the three filtering methods of Gaussian filtering,mean filtering and median filtering,the denoising effect of the scene image is compared,and the weighted mean filtering with better filtering effect is proposed,which effectively filters the high-light noise of the pipeline surface.Then,the differential image method and the projection method are used to quickly screen out the suspected damaged image,which reduces the detection workload and improves the detection efficiency.2.In order to effectively segment the damaged surface of the pipeline surface,this thesis compares the damage segmentation effect of the scene based on the commonly used first-order differential operator and second-order differential operator.The Canny algorithm for improved convolution model and the watershed segmentation algorithm based on automatic marker are proposed.In the test experiment,the Canny algorithm with the edge convolution model and the watershed algorithm with automatic mark-shadowing method can accurately segment the image damage region.3.The time of feature extraction is long and the Speeded Up Robust Features(SURF)of the scale-invariant feature transformation algorithm depend on the gradient direction of the local region pixels heavily which can result in the inaccurate feature description vector.Aiming at these problems,the SURF algorithm based on Haar-Like feature is proposed to extract features from damaged regions.Finally,the pipeline surface damage is classified using an adaptive enhancement algorithm.The experimental results show that the proposed SUAR algorithm based on Haar-Like can extract feature vectors quickly and effectively when the damaged image is blur and scale changed,further more,combined with the adaptive enhancement algorithm the classify of the damage of pipeline surface can be achieved accurately,and the recognition accuracy can reach 91%.