pdi's Introduction

PDI: Panorama Depth Image

PDI provides a solution to get the panorama depth image from a single fisheye stereo image pair. A wider view angle always benefits the environmental perception ability. For more information see https://astar.ai.

Youtube Demo Video for the SLAM using PDI.

The following steps have been tested and passed on Ubuntu 16.04.5.

1. Theoretical Background

Fisheye Camera Model: C. Mei and P. Rives, Single View Point Omnidirectional Camera Calibration From Planar Grids, ICRA 2007.

Fisheye Stereo Reconstruction: S. Li, Binocular Spherical Stereo, ITS 2009.

2. Dependencies

OpenCV: Required at leat 3.0. Tested with OpenCV 3.4.0.
OpenGL and X11

3. Run C++ Code

Compile

mkdir build && cd build
cmake ..
make

Run

./pdi

4. Run Python Code

python pdi.py

Python code does not support Operation mode.

5. Operation

5.1 'Fisheye Image' window

The first two trackbars are used to adjust the numDisparities and blockSize for OpenCV stereo matching functions. The third trackbar 'Threshold' is used to adjust the field of view of the camera.

5.2 'Panorama 3D Scene' OpenGL window

Mouse button: left drag - rotate, middle drag - pan, middle scroll - zoom, right drag - zoom.

PDI uses GLWindow library from http://ethaneade.org/.

5.3 Exit

Press 'q' or 'Esc' key on the 'Fisheye Image' window to exit.

6. Live Mode

To run pdi.cpp in a live mode, please change the variable live to true. Python code does NOT support the live mode.

bool      live = true;

and run

./pdi YOUR_CALIBRATION_FILE.yml

7. Calibration Parameter File

To run PDI in the LIVE mode, you need to download the calibration parameter file from online. Each CaliCam stereo/mono camera has a UNIQUE parameter file. Please download the corresponding parameter file by following the instructions at https://astar.ai/collections/astar-products.

pdi's People

Contributors

Stargazers

Watchers

pdi's Issues

Autoexposure

Hello. Bought your Calicam Stereo camera. In the course of working with her, two questions arose. Does it have any hardware ways to use autoexposure? If not, do you provide any code to do auto exposure?

Implementation of Kannala-Brandt camera model

Hi there,

I really like this project! Since this code only supports the Fisheye Camera Model, I was wondering if it is possible to implement the more generic Kannala-Brandt Camera Model?
I am trying to get this to work with a stereo 160 degree fisheye camera (Intel T265) but it does not seem that it can be calibrated with the Fisheye Model.
So what would have to be done (key points) to implement the Kannala-Brandt Model?

Thanks,
Tom

Recommend Projects

astar-ai / pdi Goto Github PK