Underwater Linear Object Detection Algorithm Based On Optical Imaging

Nowadays,more and more pipelines for oil and gas have been laying underwater.Oil or gas leaking will cause huge economic losses and environmental pollution once caused by equipment aging or pipeline corrosion.Research on detecting and recognizing underwater linear targets is thus of great significance for regular pipeline inspection and maintenance.Existing methods are to some extent limited on near distance,linear target detection.Based on range-gated optical system,long distance quadratic curve target detection is discussed in this thesis.The details are as follows:Firstly,High-resolution images are obtained by using the range-gated optical system.The unique characteristics of the underwater images such as low contrast,non-uniform brightness,impulse and Gaussian noises are properly dealt with contrast stretch algorithm,homomorphic filtering,median filter and wavelets.Secondly,there will appear higher ratio of false edges in traditional Canny edge detection results,which will affect the follow-up target detection step.The traditional Canny algorithm is improved by properly choosing the Sobel gradient operator,using the adaptive threshold calculation method and removing short edges in the detection result,which effectively reduces the false edge ratio.Finally,a more robust Random Sample Consensus algorithm(RANSAC)is chosen to correctly fit object's position and direction parameters from the edge image.At the same time,a pre-test step is added to the algorithm to reduce the detection time and a post correction step is added to deal with obvious error results.The final object detection time was effectively reduced without losing detection rates.