Integration and algorithm development for forward looking imaging sonars on hybrid and autonomous underwater robots

摘要

We present the integration of Blueview's P900 2D forward looking imaging sonars on two vehicles developed at WHOI's Deep Submergence Laboratory and the software to support them. For the Sentry AUV, a 6000m AUV, we developed a software system designed to run on low-power embedded Linux platforms that applies real-time computer vision techniques on acoustic returns intensity images to detect threats to vehicle safety. The system operates in the horizontal plane to simultaneously allow the collection of forward looking data that would be relevant to science. It can autonomously inform the vehicle control system to avoid threats that would otherwise result in a collision. Simultaneously, the system logs forward looking acoustic returns data for subsequent use in post processed science products. We developed a data pipeline that injects the post-processed vehicle navigation, with fused USBL and DVL navigation, in the sonar logs that the system generates and creates maps. We present results from the June 2014 R/V Atlantis expedition in the Gulf of Mexico. During this expedition we used Sentry's forward looking sonar to generate very high resolution maps of steep slope subsea escarpments sections. We also present results from our new HROV Nereid Under Ice (nUI) a 2000m depth rated fiber tethered ocean robot. First, deployed from the R/V Polarstern in a July 2014 expedition in the Arctic Ocean, nUI provided observations at the ice-sea interface. We developed software for the vehicle's Blueview P900 forward looking imaging sonar to improve under-ice precision localization and science under drifting and fragmented sea ice. Navigation under through-ice GPS-referenced poles was improved by developing software that could track the vehicle-relative position of the poles. This enabled the generation of a high-accuracy, high-frequency floe-relative vehicle position estimate. Ice floe-relative vehicle velocity, when in open water surrounded by ice floes without DVL ice lock, co- ld be estimated by applying optical flow computer vision techniques to forward looking sonar images.