ROV vision-based motion estimation: A comparison study

Different vision-based techniques used to estimate the motion of an underwater Remotely Operated Vehicle (ROV) are compared in this work. The article analyzes several different approaches both at the feature detection level and at the feature description level. In what respects feature detection, a previously used template extractor is compared with interest point detectors as the Shi-Tomasi corner detector and the Oriented FAST and Rotated BRIEF (ORB) detector. For feature description the correspondent template patch, Speeded Up Robust Features (SURF), ORB and Binary Robust Independent Elementary Features (BRIEF) descriptors are tested. All these approaches are implemented on the same free open source library allowing a fair comparison, especially in terms of computational time. The tested approaches take into account previous studies and are compared with the same batch of data collected by the Romeo ROV performing a lawn mowing pattern at constant heading and in auto-altitude mode. In estimating vehicle speed, the Shi-Tomasi corner detector combined with BRIEF descriptors and the template extractor approaches presented the lowest noise levels. In terms of computational time, template correlation outperforms all other alternatives being at least more than 2 times faster. © 2012 IFAC.