stereovisionforadas's Introduction

Stereo vision for ADAS

Stereo vision for ADAS (unsupervised learning)

Stereo matching
Stixel estimation
Stixel segmentation(clustering), object detection
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/

stereovisionforadas's Issues

Published work

Is there any published work related to it?

whether this project can be applied to our own data?

Hello, thank for your code sharing. But I wonder whether this code can be applied to our own data? Because I have seen from the file in StereoVisionforADAS that other data are not support. Could you tell the reason and what files needed to be modify to make it suitable for own data if it can? Please, thx!

Some question about principle

Hi, I am still a little puzzed by the principle of this project, so can you give me more instruction for me to modify this algorithm? As I understand, this project does as follow:

get the disparity map and remove the ground using V-disparity;
set a constant height and uses stixel world to describe the obstacle under the height;
right?
but I don't know the next step stixel segmentation, what this step based on to segment stixel into different objects?
Can you give me more instruction about this? Thx a lot ! !

Recommend Projects