Stereo vision for ADAS

Stereo vision for ADAS (unsupervised learning)

1. Stereo matching
2. Stixel estimation
3. Stixel segmentation(clustering), object detection
4. Surface Normal Vector estimation

Test on intel-core i7 desktop(cpu), processing time : 24ms/frame

Requirements

• C++ 11+
• OpenCV 2.4.9+, 3.0.0+
• OpenCV Extra (if you want to show 3D plot)
• pkg-config (mac, linux)
• cross platform(windows, osx, linux)

Usage

Input data (Test on KITTI data)

└── data
    └── left
        └── 010%d.png (start from 0000000000.png)
    └── right
        └── 010%d.png

To test a code in osx or linux

$ ./compile.sh
$ g++ -o stereovisionforadas ./*.cpp `pkg-config --cflags --libs opencv` --std=c++11
$ ./stereovisionforadas

To test a code in windows

• use visual studio, warning for opencv path (viz class)
• TBD

Algorithm

Result

Video : 3D visualization

Video : Objectness, surface normal

Acknowledgement

This project has been working in CVLab. at Inha Univ.

http://vision.inha.ac.kr/