iris MPA + Zählung
AI-Powered Passenger Counting and Occupancy Detection
THE CHALLENGE
Public transport operators need accurate passenger counts to optimize routes, allocate vehicles, and validate revenue models. Manual counting is expensive and unreliable. Legacy sensor systems struggle with occlusion, lighting changes, and crowded scenes. Existing computer vision solutions require high-end GPUs that are unsuitable for in-vehicle deployment.
THE SOLUTION
As Product Manager in the AI team at iris GmbH, I owned the development of two AI products that change how operators understand passenger flow.
Passenger Counting is an edge AI solution running YOLO-based object detection on Hailo accelerators — energy-efficient chips delivering GPU-class performance for in-vehicle use. Deployed on buses and trams, it counts passengers boarding and alighting in real time with over 95% accuracy, even in low light and dense crowds.
MPA (Multi Purpose Area Detection) recognizes occupancy of defined vehicle zones — such as multi-purpose areas for strollers and wheelchairs — with around 90% accuracy. Both systems use MQTT for low-latency messaging and comply with VDV/ITxPT standards. The data approach is GDPR-compliant: no storage of personally identifiable image data, privacy by design at every layer.
RESULTS
3 pilot projects with DACH transit operators successfully launched, market readiness achieved.