Case studies/Trafflytic
Classified vehicle counts from ordinary camera footage
Traffic studies without a road crew: cars, buses, trucks and turning movements from video a city or consultant already has.
- Client
- Trafflytic
- Location
- United States
- Year
- 2025
AI · Web
info@lyrocode.com
Input
Video
Output
Classified counts
Year
2025
Pictures from the work
Photography that matches the product — the buildings, the floor, the clothes, the road — not stock of people pointing at a laptop.
The job
Consultants and cities still send people to sit at an intersection with a clicker. The footage often already exists. They needed classified counts — light vehicles, heavies, buses — and turning movements, with a report they can put in a PDF, not a demo reel.
Weather, night, and odd camera angles break naive detectors. The product has to say when a clip is unusable, not invent a number.
What we built
Upload or RTSP in, classified counts out. We trained for vehicle classes and movement through a junction, then wrapped it in a web app: jobs, clips, QA, export. Confidence and coverage sit next to the number.
The engineering is the model and the boring parts around it — clip length, time-of-day splits, a CSV the transport planner already knows how to open.
Where it landed
A study that used to mean a crew on the kerb can start from video. Clients get class and turning counts with a trail of how the number was made. Bad footage is flagged instead of padded.
- Vehicle classification from video
- Turning-movement and time-split counts
- Job queue, QA and CSV/PDF export
- Coverage and confidence on every clip
Stack
What we shipped with — picked for this product, not for a slide.
Python
TensorFlow
FastAPI
React
Next.js
PostgreSQL
AWS
Docker
“We still do field counts when we have to. For the rest, we upload the video and get classes and turns. If a clip is junk, the product says so — that is what made us trust it.”
More work
Willis Bond·Web · Property
Project sites a development team can actually keep current
Marketing and project pages for a New Zealand developer — precinct photography, availability, and content the marketing team can change without a ticket.
Read the story
Maria B.·Commerce · Web
A fashion storefront that has to survive sale season
Commerce work for Maria B. — lookbooks, collections, and a checkout path that still holds when a drop hits.
Read the storyTell us what you are building.
We usually reply within one business day. Email info@lyrocode.com if you would rather skip the form.
