Real-time Vision-only Perception for Robotic Coral Reef Monitoring and Management

A particular challenge for Autonomous Underwater Vehicles (AUVs) operating in complex coral reef environments and conducting low-altitude reef-based monitoring and management tasks (approximately 0.5 – 1.2 m from the coral) is position estimation and obstacle detection where traditional acoustic based sensors (e.g. sonars, Doppler Velocity Logs) become less reliable due to minimum ranging distances and multipathing. Real-time stereo and monocular vision systems offer a viable means for navigation, obstacle avoidance and in-situ automated management tasks. However, such approaches are limited by their ability to robustly detect features/objects, particularly in reduced visibility conditions, false detections from dynamic objects such as fish or strong reflections/refractions from sun-glint, and motion blur. This paper provides an overview of some real-time visionbased perception approaches developed for, and applied to, small-scale AUVs for improving underwater navigation and obstacle avoidance in coral reef environments. We also present applications of our latest work for AUV-based coral reef management tasks including control of crown-of-thorns starfish and coral larvae reseeding over damaged reefs.