Maritime EO/IR object detection for autonomous vessels.

Challenge

Detecting vessels, navigation aids and hazards from electro-optical / infrared sensors at sea, where glare, spray, horizon ambiguity and tiny far-field objects make labeling hard and error-prone.

Approach

io-ai annotated primary objects (vessels, buoys & aids-to-navigation, shoreline polygons, fixed structures, birds) plus extended targets: track IDs, oriented vessel boxes, people on deck or in water, debris, and an ownship mask.

Rich scene-level metadata was captured separately: horizon line, ignore regions for spray and glare, time-of-day and weather, visibility band in nautical miles, and Beaufort sea state (0–9). Strict rules applied: tight boxes, a 7×7px minimum, occlusion flags, never label spray, glare or wakes as objects, and prefer “unknown” over a wrong guess.

Quality

Every task ran through io-ai's multi-level QA pipeline, with first-batch audits before scale-up and full rejection-reason metadata on every item, holding over 98% accuracy on rolling audits, with weekly status reporting.

Result

A clean maritime perception dataset with scene context that supports detection in degraded visibility, the kind of edge-case rigor most generalist vendors don't offer.