Optical Flow Visualization

This project implements and visualizes optical flow using the Lucas-Kanade method. Optical flow is a computer vision technique that estimates motion patterns in image sequences, providing valuable information about how objects move within a scene.

Features

• Real-time optical flow visualization from webcam input
• Color-coded flow vectors where color represents direction and intensity represents velocity magnitude
• Customizable parameters for smoothing, stride, and image downsampling
• Support for both image files and webcam input

How It Works

The implementation follows these key steps:

1. Derivative Calculation: Computes spatial and temporal image derivatives
2. Structure Matrix Construction: Builds time-structure matrices from derivatives
3. Velocity Estimation: Solves the optical flow equation for motion vectors
4. Visualization: Renders motion vectors with direction-based color mapping

Requirements

• Python 3.x
• OpenCV (cv2)
• NumPy

Installation

# Clone this repository
git clone https://github.com/arda92a/Optical-Flow.git
cd optical-flow

# Install required packages
pip install opencv-python numpy

Usage

Webcam Demo

Run the webcam demo to see optical flow in real-time:

python optical_flow.py --webcam --smooth 3 --stride 1 --div 1

Parameters:

• --smooth (default: 3): Window size for smoothing the structure matrix
• --stride (default: 1): Downsampling factor for velocity calculation
• --div (default: 1): Downsampling factor for input images

Image Processing

Process a pair of images to calculate optical flow:

python optical_flow.py --image1 first_image.jpg --image2 second_image.jpg --output flow_result.jpg

Examples

Input Images and Flow Result

The optical flow algorithm processes sequential frames to detect motion. Below are sample results:

First input frame (starting point)

Second input frame (ending point)

Resulting optical flow visualization showing motion vectors

Movement Experiments

Below is a comprehensive visualization of different movement patterns and their corresponding optical flow representations:

Various movement patterns and their optical flow signatures

This visualization demonstrates how different types of motion (translation, rotation, scaling, etc.) appear in the optical flow field, providing a reference for interpreting flow patterns in real-world applications.

Implementation Details

• Box filtering with integral images for efficient smoothing
• Direction-colored visualization using HSV color mapping
• Aperture problem handling for areas with insufficient texture
• Velocity constraint to avoid extreme flow vectors

Advanced Options

You can adjust several parameters to balance accuracy and performance:

# Higher smoothing for noisy images
python optical_flow.py --webcam --smooth 5

# Increase stride for faster processing
python optical_flow.py --webcam --stride 2

# Decrease resolution for better performance
python optical_flow.py --webcam --div 2

Troubleshooting

• Slow performance: Increase the stride and div parameters
• Noisy flow vectors: Increase the smooth parameter
• Webcam not detected: Check your camera connection and permissions

Credits

This implementation is based on the Lucas-Kanade method for optical flow estimation.