AI airport perception system
Mishandled baggage is one of the most expensive failure modes in commercial aviation. An international carrier wanted the airport itself to see the problem as it happened.
Problem
Airport baggage operations span conveyors, ramps, loaders, and aircraft — many actors, many hand-offs, many places a bag can go wrong. Traditional tracking relies on discrete scans and paper exceptions. The R&D team at an international carrier wanted a real-time perception layer that could watch the environment itself, follow individual bags across sensors, and surface mishandling in the moment rather than as a post-event complaint.
Approach
Designed and deployed an advanced perception system for the airline’s R&D team. Combined spatial depth cameras positioned across baggage operations with both edge and cloud image recognition, plus a multi-object tracking layer that maintains identity across camera hand-offs. Prototyped real-time baggage tracking and mishandling detection as the first use case — the harder problem under the hood is holding a stable track on a moving object across multiple sensors and operators.
Outcome
- Working prototype of real-time baggage tracking deployed for the carrier’s R&D team
- Mishandling detection demonstrated end-to-end on live baggage operations data
- Reference architecture for multi-camera edge/cloud perception reusable in other airline workflows
- TODO: verify metric — tracking accuracy / ID-switch rate
- TODO: verify metric — mishandling detection precision / recall