Underwater Object Detection – Remote Control Interface

A unified control and monitoring interface for an underwater object detection drone, developed as part of the IIT Guwahati internship project (July 2025).

---

Overview

This system provides a cross-platform GUI to interact with a remotely operated underwater drone equipped with:

• Intel RealSense camera
• NVIDIA Jetson Nano for ML inference
• YOLOv8 object detection model

The interface supports real-time control, recording, playback, and Python-based ML workflows, accessible via a modern frontend.

---

Features

• Live video feed from Jetson or webcam
• Start/Stop recording sessions
• Save videos as .mp4 with timestamped filenames
• Display and playback saved recordings
• Trigger and display YOLO-based object detection from the frontend
• Frontend/backend integration (Vite + Flask)
• Supports remote backend running on a Jetson Nano via odstream-remote

---

Tech Stack

Area	Tools
Frontend	React (via Vite), TailwindCSS
Backend	Python, Flask
Remote Backend	Flask on Jetson Nano
Object Detection	YOLOv8 (Ultralytics)
Hardware Targets	Jetson Nano, RealSense Camera

---

Setup

See git & workflow instructions here

---

Directory Structure (WIP)

odstream
├── .gitignore           # Self explanatory
├── docs                 # Dev logs, documentation
│   └── git-setup.md     # Project setup and git guide
├── shared
│   └── recordings       # Saved .mp4 files
├── backend
│   ├── utils            # Python helpers (planned)
│   ├── requirements.txt # Backend dependencies
│   └── app.py           # Flask entrypoint
└── frontend
    ├── public           # Favicon, other global assets
    └── src              # Everything app related
        ├── pages        # Route-based page components
        ├── utils        # Helpers functions (planned)
        ├── assets       # Assets for components
        └── components   # Reusable UI components
            └── ui       # UI primitives

# Jetson-side remote backend (separate repo)
odstream-remote
├── .gitignore           # As per usual
├── app.py               # Jetson Flask entrypoint
├── jetson_camera.py     # Stream handling and detection logic
├── jetson_routes.py     # Defines all Flask API endpoints
└── requirements.txt     # Dependencies for remote backend

---

Roadmap

Phase 1 – Web App (Vite + Flask)

• Vite + Tailwind frontend
• Video source selection dropdown
• Live camera feed display
• Recording and saving .mp4 files
• Playback for saved videos
• Stream from Jetson Nano
• Jetson connection settings in modal

Phase 2 – Detection Pipeline

• Flask API to run YOLOv8 inference
• Detection overlay on live stream
• Toggle detection from frontend

Phase 3 – Deployment & Polish

• Auto-deploy scripts for Jetson + host
• LAN streaming support (MJPEG)
• Basic mobile responsiveness

---

Contributors

• Poran Dip (primary developer)
• Ruhan Roushan Islam (multi-model integration)
• Pooja Basumatary (testing & presentation)

Mentors:

• Divya Jyoti (PhD Scholar, EEE, IIT Guwahati)
• Parvez Aziz Boruah (PhD Scholar, CSE, IIT Guwahati)

Supervisor:

• Dr. Chayan Bhawal (Professor, EEE, IIT Guwahati)

---

Notes

• Minimal dependencies, modern stack
• Recording format: YYYYMMDD-HHMM.mp4 (e.g., 20250704-1315.mp4)
• Network-based live stream and object detection are now supported

---

“Build it simple, scale it smart.”