aujfa / stereo-vision-3D-reconstruction

A C++ application for 3D image reconstruction from two images using stereo vision technique

Stereo Vision 3D Reconstruction

This is a C++ application for 3D image reconstruction from two images using stereo vision technique. Stereo pair was obtained with two low resolution web cameras.

stereo-vision-3D-reconstruction.mp4

Requirements

• Two webcams or stereo camera
• openCV 3.4.0 (with contrib module)
• A chessboard pattern pattern.png which you can download from here

Steps

1. Print the chessboard pattern and paste it on a hard surface or display it on your phone or tablet. The print/image needs to be as straight as possible!
2. Capture multiple images of the chessboard pattern from the different views (as you can see here) for calibration using takeChessboardImages() function from camera_calibrator.cpp.
3. I have included one stereo pair. You need to make your own with the cameras you calibrated. Create a stereo pair using takeStereoPair() from camera_calibrator.cpp.
4. Finally, reconstruct the scene with ReconstructPointCloud() from camera_calibrator.cpp.

* blendPictures() is a helper function to ease the process of aligning the cameras for the stereo pair capture. It is not needed for reconstruction.

* opencv_world340d.dll is compiled binary for Visual Studio 2017 32-bit. If you have some other IDE, you need to compile your own binary from source.

Resources

• Most of the theory behind the code is explained in 3-D Reconstruction with Vision by Venkatesh Tata
• Article about Viz from openCV contrib module: Viz - New 3D visualization module in the OpenCV library
• Instruction how to make your own binary: Build opencv 3.4 under windows 10 with contrib library, Git Source ,CMAKE ,Visual Studio 2017
• If you are interested in details about matrices for calibration, triangulation, etc. there is a great lecture by Cyrill Stachniss: Photogrammetry