Stereo Matching PRO

Dive into the world of computer vision with this comprehensive Python project that brings stereo matching to life. Designed for enthusiasts and professionals alike, this project leverages cutting-edge algorithms to generate precise depth maps from stereo image pairs. Whether you're exploring autonomous systems, 3D reconstruction, or just expanding your knowledge, this project offers a hands-on approach to mastering stereo vision.

Table of Contents

• Environment setup
• Data
• Running
• Reference

Environment setup

This code was tested with Python 3.8.3.

Installation

git clone https://github.com/june65/StereoMatchingPro
pip install -r requirements.txt

Data

Middlebury Computer Vision Pages provide datasets and benchmarks for computer vision research, including stereo vision, optical flow, and multi-view stereo. Download datasets at link here.

Running

We demonstrate the precision of our stereo matching algorithms using the benchmark Tsukuba dataset.

Ground Truth

Absolute intensity difference

python main.py --costmethod AD --rgbexpand True

Squared intensity difference

python main.py --costmethod SD --rgbexpand True

Sum of absolute differences

python main.py --costmethod SAD --costwindow 3

python test.py --costmethod SAD --costwindow 33

Sum of squared differences

python main.py --costmethod SSD --costwindow 3

python test.py --costmethod SSD --costwindow 33

Adaptive Support Weights

python main.py --costmethod ASW --costwindow 33

python test.py --costmethod ASW --costwindow 51

Semi-Global Matching

python main.py --costmethod SGM

Graph Cut

python main.py --costmethod AD --graphcut True

Left Right Consistency Check

python main.py --costmethod SAD --costwindow 3 --lrcheck True

Tree Filtering

python main.py --costmethod ASW --costwindow 27 --treefilter True --lrcheck True

Weighted Median Filter

python main.py --costmethod ASW --costwindow 33 --midfilter 5

Best Result

ASW + Left Right Consistency Check + Tree Filtering + Weighted Median Filter

python main.py --costmethod ASW --costwindow 33 --lrcheck True --treefilter True --midfilter 5 

Reference

Stereo Matching With Fusing Adaptive Support Weights

Stereo Processing by Semi-Global Matching and Mutual Information

Stereo Matching Using Tree Filtering

Constant Time Weighted Median Filtering for Stereo Matching