Navigation Vision

This repository contains a set of Python scripts for object detection and 3D localization using a monocular camera. The project uses a YOLO model (for example, my_coin.pt) along with additional modules to:

• Detect objects and draw bounding boxes.
• Estimate distance (range) and horizontal bearing (angle) from the camera to the object using the pinhole camera model.
• (Optionally) Compute a 3D coordinate estimate by combining YOLO detections with a monocular depth estimation model (MiDaS).
• Record the detection video including overlays such as bounding boxes, distance, bearing, and 3D coordinates.

Note:

• The distance estimation requires calibration of camera parameters (e.g., focal length, known object width).
• The MiDaS depth output is provided in relative units. For accurate depth (e.g., in centimeters), you may need to calibrate the output.
• Large files (such as demo videos) are not included in the Git history.

Table of Contents

• Prerequisites
• Installation
• Run
• Project Structure

Prerequisites

• Python 3.7+
• pip
• Git

You will also need to install the following Python packages:

• torch
• opencv-python
• numpy
• Pillow
• timm
• ultralytics

Installation

 pip install torch opencv-python numpy Pillow timm ultralytics
 pip install timm

1. Clone the Repository

   git clone https://github.com/Cyclone-Robosub/Navigation-Vision.git
   cd Navigation-Vision

Run

python my_coin_script.py --model "my_coin.pt" --source "usb0" --resolution "640x480" --record

Project Structure

• my_coin_script.py Main script for object detection and 3D localization. It overlays bounding boxes, distance, bearing, and estimated 3D coordinates on the camera feed and can record the output video when the --record flag is used.

• distance_estimator.py Contains functions for distance and bearing estimation using the pinhole camera model.

• midas_depth_estimator.py Loads and runs the MiDaS model to generate a depth map from an image frame.

• video_recorder.py A standalone script to record raw video from your camera.

• my_coin.pt Your trained YOLO model file.

Other files: Readme.txt, sample images, videos, etc.