Data 2024–2026 · Ongoing · Client: Dr. Humbert Salvadó — Zoology group, Universitat de Barcelona (UB) · Responsible: Francesc Xavier Gómez
The challenge
The zoology group of Dr. Humbert Salvadó (Universitat de Barcelona) needed to study the swimming behaviour of Paramecium aurelia — a model ciliate for motility and chemotaxis — in long, unattended sessions. Manual tracking is slow and inconsistent, and off-the-shelf solutions could not keep a fast, elongated organism centred on a motorised stage in real time at a research-lab budget.
The solution
Winkoms built a closed-loop, real-time tracking system on its own motorised SimpleXYZ stage. A YOLOv8-segmentation model — running on an Axelera Metis M.2 accelerator on a low-cost Radxa ROCK 5 ITX+ — detects the organism frame by frame; a QR code overlaid on each video frame carries the stage position in microns, so the software converts the detected centroid into real-world coordinates and drives the XY stage to keep the cell centred. The pipeline exports velocity, displacement and trajectory for quantitative analysis,.
The results
The system reaches 88.6 FPS and stays thermally stable at 38 °C, sustaining a continuous tracking session of almost 6 hours. Position jitter is 17.7 µm (7% of the organism’s body length) and the recovered kinematics match published P. aurelia data. The project was selected winner of the Axelera AI “Demo Jam #1”, recognising a fully deployed, real-world use of the Metis M.2.
Read Axelera’s official announcement →