Robot Vision Demo Using DepthAnything

A simple demonstration of using DepthAnything for depth-based robot navigation. This demo shows how to process real-time camera feed, estimate depth information, and generate structured navigation commands for a robot based on depth analysis.

Features

• Real-time camera feed processing
• Depth estimation using DepthAnything model
• Region-based depth analysis for navigation
• Object detection based on depth gradients
• Automatic navigation command generation
• Web-based visualization interface
• Real-time depth visualization

Depth Analysis

The system uses DepthAnything to provide detailed depth information:

Region Analysis

• Divides the image into 5 regions (center, top, bottom, left, right)
• Provides depth statistics for each region:
  • Mean depth
  • Minimum depth
  • Maximum depth
  • Relative distance classification (near/far)

Object Detection

• Detects objects based on depth gradients
• For each object provides:
  • Position (normalized x, y coordinates)
  • Size (width and height as proportion of image)
  • Depth information (mean depth and relative distance)

Visualization

• Generates depth map visualization
• Shows region divisions
• Highlights detected objects

Installation

1. Create a virtual environment:

python -m venv venv
source venv/Scripts/activate

2. Install requirements:

pip install -r requirements.txt

Usage

Main Application

1. Start the server:

python app.py

2. Open your web browser to http://localhost:53549/

3. Allow camera access when prompted

4. Click "Start Processing" to begin real-time processing

Depth Analysis Interface

1. Start the depth analysis interface:

python web_interface.py

2. Open your web browser to http://localhost:53549/

3. Use the interface to:

   • Upload images for analysis
   • View depth maps and visualizations
   • See region-based depth analysis
   • Inspect detected objects

Command Line Usage

You can also use the depth analyzer directly from the command line:

python depth_analyzer.py path/to/image.jpg

This will:

• Generate a depth analysis
• Save visualization to 'depth_analysis.png'
• Print detailed analysis results

Command Structure

The system generates JSON commands in the following format:

{
    "velocity_command": {
        "linear_velocity_mps": 0.5,    // Forward/backward speed (-1.0 to 1.0)
        "angular_velocity_radps": 0.2   // Turning speed (-1.0 to 1.0)
    },
    "gait_mode": "trotting",           // Robot's movement style
    "reasoning": "Moving forward to approach the target object while avoiding the obstacle on the left",
    "timestamp": "2024-01-31T17:00:03Z"
}