Skip to content
/ rgbdrec Public

Depth camera pose estimation utility, with common abstraction on top of COLMAP, ORB_SLAM2

Notifications You must be signed in to change notification settings

sxyu/rgbdrec

Repository files navigation

Overview

This is a common interface for registering point clouds captured using a depth camera (Kinect V2 or Azure Kinect), supporting COLMAP and SLAM-based solutions. A general data capturing program is included.

Binaries

  • rgbdrec: Basic RGB-D sequence recorder
  • rgbdreg-colmap, rgbdreg-orbslam2: Point cloud registration (via odometry/SFM): supports COLMAP and ORB_SLAM2. Uses data format output by rgbdrec
    • rgbdreg-colmap: registration method based on COLMAP. Note the SFM is performed using color images only and is initially in an arbitrary unit; we transform all matched feature points to camera space and compare the depth to the depth image to correct the scale. Requires colmap to be in system PATH.
    • rgbdreg-orbslam2: registration method based on ORB_SLAM2. This method actually uses RGB-D information of each frame and produces correctly scaled camera poses.
  • rgbdreg-viewer: Visualize camera pose estimation. After running one of the above, use rgbdreg-viewer <data-folder> to ensure result is good.

Screenshot of registered point cloud

Format

Each registration program produce a file poses.txt (readable with numpy.loadtxt) in the data directory. This is a text file of floats, separated by spaces and newlines, of shape [N, 12] where N is the number of images; each row can be viewed as a [3, 4] row-major rigid-body transform matrix (bottom row omitted) which transforms a point in homogeneous coordinates from camera space to world space. As usual, the 3 leftmost columns are the rotation and right column is the translation.

Regardless of program used, the output coordinate system will be right-handed with axis orientations [x, y, z] = [right, up, backward (screen to your face)].

See rgbdreg-viewer.cpp for an example program reading poses.txt.

Setup

Dependencies

  • OpenCV 3+
  • Eigen 3
  • Boost (really I only use program_options, filesystem, process)

Optional Dependencies

Build

Build using cmake as usual: mkdir build && cd build && cmake .. && make -j12

Credit

Out of laziness I reuse a lot of code from my other project https://github.com/sxyu/avatar, which also borrows from https://github.com/augcog/OpenARK

About

Depth camera pose estimation utility, with common abstraction on top of COLMAP, ORB_SLAM2

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published