Automated visual servoing for close inspection using low-cost, man-portable vehicles

Remotely Operated Vehicles (ROVs) now offer sonar-based control that enable autonomous inspection capabilities at stand-off distances greater than the blanking range of the sonar. At near ranges, equivalent autonomy behavior must be provided using on-board video sensors. This paper presents a sensor-based servo capability where a suite of onboard Automatic Target Recognition (ATR) systems fuse input from multiple sensors (sonar and video) to provide the required input to the control system, allowing the ROV to maintain station relative to an object of interest. The video ATR is then presented; this is trained in-mission, leveraging the ROV user interface and high communication bandwidth to allow the operator to select and provide training data to the ATR. The system allows the ROV autonomy to be deployed against a priori unknown targets selected by the pilot. The video ATR uses a Fern Classifier in conjunction with an Optical Flow based tracker to allow the ATR to cope with variations in the target appearance over time. SeeByte´s ROV Dynamic Positioning product, SeeTrack Copilot, is then presented before discussing the integration of these technologies to provide a novel, end-to-end solution for ROV auto-transit, dynamic positioning and visual servoing. Results are presented using the VideoRay Pro4 micro-ROV platform. The demonstrated prototype technology offers potential as a diver replacement in applications as varied as Mine Counter Measures and Underwater Archeology.