nbfigueroa / 6d-rgbd-odometry Goto Github PK

This package provides a novel 6D RGB-D odometry approach that finds the relative camera pose between consecutive RGB-D frames by keypoint extraction and feature matching both on the RGB and depth image planes.

To find the Rigid Motion between two camera frames, the VO algorithm follows these steps:

#####1) RGB-D Keypoint Extraction:

STAR Detector for RGB image and NARF detector for Depth Image.

#####2) RGB-D Feature Matching:

BRAND Description computation on keypoints and brute-force descriptor matcher with the Hamming norm and correspondence cross checking.

#####3) Correspondences Outlier Rejection:

Filter the matches using RANSAC outlier rejection algorithm

#####4) 6D Rigid Motion Estimation:

Iterative estimation using RANSAC and Umeyama Method.

The output of this algorithm should be something similar to this video:

To use this code, you will need to install PCL library (www.pointclouds.org), OpenCV (www.opencv.org) and compile the brand library in this folder, to do so, (after succesfully installing PCL and OpenCV) you must follow these instructions:

cd ~/brand/
mkdir build && cd build
make
sudo make install

Once compiled you can test the BRAND desciptors by running a simple matching example found in:

###brand_match This stand-alone example shows how to use the class to create BRAND descriptors and a simple brute-force matching between two frames.

Compiling:

cd ~/brand_match/
mkdir build && cd build
make

Running

./brand_match_demo ../../test_images/real-poseA-t01-rgb.png ../../test_images/real-poseA-t01-depth.png ../../test_images/real-poseA-t02-rgb.png ../../test_images/real-poseA-t02-depth.png  525 525 319 239

Where, fx = fy = 525, cx = 319, cy = 239 (i.e. the kinect intrinsic parameters)

You should then get the following output:

###brand_odometry Finally, if you want to test the odometry algorithm, you should download a dataset with a moving rgb-d sensor. I recommend one of these : TUM RGB-D Dataset or if you have your own dataset you should just put both rgb and depth in the same folder with same timestamps, follow the naming convention of the rgb-d dataset.

** Not working, missing a header file **

Compiling:

cd ~/brand_odometry/
mkdir build && cd build
make

Running:

./brand_odometry ${path-to-dataset} 

#####Reference: Nadia Figueroa, Haiwei Dong, and Abdulmotaleb El Saddik. 2015. A Combined Approach Toward Consistent Reconstructions of Indoor Spaces Based on 6D RGB-D Odometry and KinectFusion. ACM Trans. Intell. Syst. Technol. 6, 2, Article 14 (March 2015), 10 pages.

H. Dong, N. Figueroa and A. El Saddik, "Towards consistent reconstructions of indoor spaces based on 6D RGB-D odometry and KinectFusion," 2014 IEEE/RSJ International Conference on Intelligent Robots and Systems, Chicago, IL, 2014, pp. 1796-1803.

E. R. Nascimento, G. L. Oliveira, M. F. M. Campos, A. W. Vieira and W. R. Schwartz, "BRAND: A robust appearance and depth descriptor for RGB-D images," 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, Vilamoura, 2012, pp. 1720-1726.

Brand Code provided by: Erickson R. Nascimento 6D-RGD-Odometry Code provided by Nadia Figueroa.