Ali Akpoyraz
← Work

VELARİS.

AI video analysis·2026·Reached the Final Delivery Report stage

A video analysis pipeline for the TEKNOFEST 2026 Smart Road Safety with 5G and AI competition that reads the vehicle, the plate, the driver’s behaviour and the passengers from traffic footage.

Source code ↗

What was there

Each team had to build a system that takes a traffic video and reports the vehicle’s type, colour and plate, the driver’s violations and the passengers, each with a timestamp.

The output had to match a fixed JSON schema read by the scoring script exactly; a single malformed record could break the scoring.

How I built it

One model per task: vehicle type, colour, plate, driver action, objects and passengers. Changing one does not touch the others.

Swerving is not a separate model; it is computed from the trajectory of the vehicle’s position frame by frame.

A one-off detection on a single frame never reaches the report; a finding has to show up across several nearby frames first.

Before writing, the output is cleaned against the schema: unknown labels and categories are dropped, plate format and confidence scores are validated.

The pipeline is packaged in a GPU-enabled Docker container so it runs the same way in the competition environment.

What I did

  • Merging the models trained by the team into a single inference pipeline.
  • Swerving detection from the vehicle’s trajectory.
  • Chaining plate detection and EasyOCR for plate reading.
  • JSON output that matches the scoring schema, with validation tests.
  • The Docker package and a delivery layout that follows the Final Delivery Report.

Project details

Client
TEKNOFEST 2026 team project
My part
Team member: model integration, swerving detection, container
When
Built in 2026
Tools I used
Python · YOLO (Ultralytics) · EasyOCR · OpenCV · Docker
Next projectmalatyakuyumcu.com →

I can do the same for you.

If you need something close to this project, write and we will look at it together.

Let's talk →