Dmytro Honcharenko

I estimated a go-kart’s yaw rate from onboard video alone — no sensors.

Seeking a 2027 alternance in computer vision / ML.

FastTrack

A research internship at SERLI: can a single onboard camera replace the sensors that record a kart's telemetry? My part was the yaw rate — how fast the kart turns, frame by frame — the number a coach uses to tell whether a driver turned in early or late.

The pipeline matches features between consecutive frames (SuperPoint + LightGlue) and recovers the rotation geometrically. Against Assetto Corsa simulator ground truth it reaches r = 0.934 and a mean error of 3.20°/s. Validated in the simulator only — real karting footage had no telemetry to check against.

Line chart comparing the median-filtered yaw-rate estimate to simulator telemetry over an eighteen-minute session; the two lines track closely throughout.
Estimated yaw rate vs. simulator telemetry over an 18-minute session.
  1. Full internship report — French, 20 pp (English abstract, p. 1)
  2. Reference implementation — extracted from the original project, not runnable standalone

Engineering background

Three years of freelance engineering, Binaryx as the main client — full-stack web and mobile development, CRM integration, and cloud delivery, before turning to computer vision.

React · TypeScript · Next.js · Nest.js · PostgreSQL · Swift · SwiftUI · React Native · AWS

Education

  1. Master, Computer Science – Université de Poitiers, France · 2026–2028
  2. Licence, Computer Science – Université de Poitiers, France · 2023–2026
  3. Junior specialist – Cherkasy State Business College, Ukraine · 2018–2022

Beyond engineering

Linguistics · International law · Formula 1

Contact