Skip to content
FH
All projects
Academic project

Computer Vision · AI/ML

Smart Camera Lobster

Real-time underwater lobster detection and counting

Computer-vision system that enhances underwater footage, detects and counts lobsters, and tracks them across frames for remote monitoring.

YOLOv7StrongSORTPythonOpenCVJetson NanoJetson Xavier NX
Lobster detection output

Overview

A team final project built to help Aruna monitor and count lobsters remotely, reducing dependence on divers and weather conditions. The pipeline enhances underwater images, detects and counts lobsters, and tracks individual animals across frames. It was tested on edge devices and a laptop.

Problem & solution

The problem

Manual lobster monitoring relied on divers and favourable weather, which made counts expensive, infrequent, and difficult to verify over time.

What was built

An end-to-end vision pipeline that runs on edge hardware near the farm: enhance the frame, detect and count lobsters, then track them so the same animal is not counted twice.

Key features

  • Underwater image enhancement before inference
  • Lobster detection and counting with YOLOv7
  • Object tracking across frames with StrongSORT
  • Deployment and testing on Jetson Nano, Xavier NX, and laptop hardware
  • Reported team results: 0.996 mAP@0.5 and 14 FPS on laptop/GPU

My contribution

  • Team final project (2023). Individual contribution is not documented in the available materials.

Visual showcase

Lobster tracking output visual
Tracking output
Underwater image enhancement visual
Underwater enhancement

Related projects

View all