An Efficient Underwater Pipeline Detection System Using Machine Learning Approach

Underwater object detection is very essential for pipeline detection, tracking and safe navigation. This paper presents methods to detect pipeline by using Support Vector Machine and Random Forest classifiers with 5 features - Histogram of Oriented Gradients, Scale Invariant Feature Transform, Speed-UP Robust Features, Shi-Tomasi Corner Detector and Harris Corner Detector from the images taken by Remotely Operated Vehicle. Among these techniques, Random Forest classifier with Histogram of Oriented Gradients gives the best accuracy of 98.81%.