Background
In theme parks, speed isn't optional. Roller-coaster riders cycle through by the second, water slides produce hundreds of triggers per hour, and every single photo has to land at the self-service kiosk within seconds — properly framed, post-processed, ready to buy or download. Standard photo workflows collapse under that mix of volume, latency and harsh outdoor conditions.
Plutos Company UG came in with a clear picture: a fully automated system that controls the professional camera, picks the right frame from each run, crops it correctly, applies the park's branding and serves the result directly to the kiosk. International by design, robust against salt air, vibration and staff turnover.
Solution
We built the entire pipeline — from the screw on the mainboard to the kiosk front-end.
- Hardware-level triggering. Raspberry Pis sit directly at the ride, synchronised with the professional camera trigger. Sensors and PLC interfaces are wired in so that hitting the right second is configuration, not luck.
- Detection with custom YOLO models. A Python microservice runs YOLO networks trained on the actual data of each park attraction. It reliably detects the frame where riders are actually visible — even under extreme speeds, lighting changes and water spray. Models were refined iteratively on real-world footage from each site.
- Image processing as a pipeline. Detected frame → automated crop → corrections → park branding → output format. Every stage is observable, swappable, reproducible. New attractions with different setups are rolled out by configuration, not by code forks.
- Kiosk front-end. Riders find their photo via code, wristband or selection and decide whether to buy or download. Payment integrates with the park's own POS systems.
Outcome
The Plutos pipeline has been running across four countries in parallel for several years — Vietnam, Finland, Germany and the USA. Hundreds of thousands of photos are processed automatically every month, with no deep technical know-how needed at the park. When a park onboards a new ride, the first photos typically run through the system the same day.
What started as "a cool photo idea" has turned into a product that competes internationally in professional theme-park photography — built by a small, focused engineering team.
Why this fits us
This is not a classic web project. Not a classic AI project. It lives in the overlap: precise hardware integration, robust real-time computer vision, well-structured data pipelines and a front-end that needs to be operated by park staff. We enjoy building systems like this — when the bar is that money reliably moves through a kiosk in a park in Hanoi, and not just through a slide deck.