About the Multi-Resolution Stereo Analysis Project
Motivation: This project aimed to implement a multi-resolution stereo analysis system leveraging region and feature-based matching to enhance depth perception and disparity calculations for stereo image pairs. The focus was on achieving accurate disparity maps by integrating advanced algorithms with customizable user options.
Techniques Used
Region-Based Analysis
- User-defined template and matching window sizes (square or rectangular).
- Matching performed along the x-direction using metrics such as:
- Sum of Absolute Differences (SAD)
- Sum of Squared Differences (SSD)
- Normalized Cross-Correlation (NCC)
Feature-Based Analysis
- Harris corner detector for feature extraction and matching.
- Matching scores included SAD, SSD, and NCC for flexibility.
Validity Check and Gap Filling
- Disparity validation through left-to-right and right-to-left matching.
- Gaps filled using adaptive neighborhood averaging with options for larger window sizes for robust interpolation.
Performance
- Successfully generated accurate disparity maps for stereo pairs.
- Incorporated region-based analysis at the top resolution and feature-based matching at lower resolutions for optimal results.
Results
- Enhanced depth perception for 5 different stereo pairs.
- Validated multi-resolution analysis with disparity maps across varying levels.
- Minimized disparity errors and improved stereo image alignment accuracy.
Future Scope
- Enhanced Features: Integration with deep learning-based matching methods.
- Faster Algorithms: Real-time disparity calculations using parallel processing.
- Applications: Deployment in autonomous systems and 3D reconstruction tasks.
GitHub Repository
Explore the details and code for the Multi-Resolution Stereo Analysis Project: Multi-Resolution Stereo Analysis Project Repository