Computer vision on NVIDIA Jetson Nano

Edge AI Livestock Monitoring

A real-time edge vision system combining detection, segmentation, counting and motion tracking on live video.

Period
Spring 2025
Team
TUBITAK 2209-A university research project
Role
Team lead and system design
Technologies
  • Python
  • YOLOv8
  • SAM2
  • OpenCV
  • NVIDIA Jetson Nano
  • Linux

Key contributions

  • Led system design for a TUBITAK 2209-A research project.
  • Implemented YOLOv8 detection and SAM2 segmentation on live video.
  • Built OpenCV pipelines for animal counting and motion tracking and authored the project proposal.

Problem

Livestock monitoring needs useful signals from continuous video while operating on an edge device rather than relying on an unspecified cloud pipeline.

Context and responsibility

Ayhan led the system design for a TUBITAK 2209-A University Students Research Projects Support Program submission and authored the proposal.

Architecture

Live video passed through OpenCV pipelines. YOLOv8 handled object detection and SAM2 handled segmentation. Downstream logic performed animal counting and motion tracking on Ubuntu/Linux running on an NVIDIA Jetson Nano.

Important decisions

  • Combine detection and segmentation rather than relying on a single vision output.
  • Run the vision pipeline on an explicit edge-computing target.
  • Connect perception outputs to domain signals: counting and movement.

Evaluation and production constraints

The hardware target and algorithms are verified. Frame rate, accuracy, power use, thermal behavior and field-test results are not publicly documented.

Outcome

The system design and proposal were delivered under the TUBITAK 2209-A program.

Back to projects