Continuous real-time or near-real-time species identification and automatic counting in marine imagery is increasingly being viewed as a future method for managing the marine environment. Currently, automated detection and classification require imagery to be stored and then processed in a lab-based, cloud system on land. Edge computing and processing of marine imagery in the field is the next step in obviating the need for storing and moving large amounts of data ashore. The last step is integrating the system into an autonomous vehicle to perform continuous monitoring of relevant areas.
We are currently developing an edge detection system for seabirds on RV Investigator as part of the CSIRO led South East Australia Marine Ecosystem Survey (SEA-MES). This could allow deployment to a range of crewed vessels including commercial fishing vessels. A system deployed to an autonomous vehicle however would be more flexible, and cost-effective, and offer a means of collecting data for offshore wind developments and marine environmental regulators. This talk will outline our current plan and progress in a project that is developing an edge AI system integrated into an autonomous Maritime Robotics Otter, surface vehicle.