Powered by GitBook

Real Time 3D Reconstruction from Monocular Video

January

1st week

  • Still at Christmas's holidays!

2nd week

  • Importing lsd_slam code into QtCreator for better code navigation and understanding
  • Breaking lsd_slam module into small reusable pieces of code for a semi-dense map reconstruction
  • Removing ROS dependencies from the code.
  • Adjusting ATAN camera calibration computed by PTAM using the following parameters: [ 0.507403 0.731726 0.449718 1.00829 0.0278688 ]
  • Created a little roslaunch file to automate the execution of the mapper and viewer in a single command:
<launch>
<node name="webcam" pkg="usb_cam" type="usb_cam_node" output="screen" >
<param name="video_device" value="/dev/video0" />
<param name="image_width" value="640" />
<param name="image_height" value="480" />
<param name="pixel_format" value="mjpeg" />
<param name="camera_frame_id" value="usb_cam" />
<param name="io_method" value="mmap"/>
</node>
<node name="mapper" pkg="lsd_slam_core" type="live_slam" respawn="false" output="screen">
<remap from="image" to="/usb_cam/image_raw" />
<param name="calib" value="creative_hd_webcam.cfg" />
</node>
<node name="viewer" pkg="lsd_slam_viewer" type="viewer" respawn="false" output="screen" />
</launch>

3rd week

  • Fixed the issue at compiling with catkin support thanks to the patch in CMakeList.txt and the new folder layout of the refine-devel branch of a fork of lsd_slam
  • Recompile image_pipeline avoiding unnecessary GTK2 dependencies which mix symbols with GTK3
  • QGLViewer is recompiled with support for Qt5 (the version in the Ubuntu repositories is still dependent of Qt4)
    • Still having issues in the function qRegisterResourceData during Qt5 initialization
    • Moving again to origin remote version of lsd_slam_viewer, but using the patched version of CMakeLists.txt from the forked repository

  • Issue with generated point clouds: degenerates in camera vision ray, probably due to camera noise at computing depth map

    bad-pointcloud1 bad-pointcloud2

  • Recalibrate webcam with smaller (so better) mean reprojection error than before (0.33), but a different estimated focal length (766.44 pixels)

    • Focal length in mm can be computed as $f = fx / mx$, where $mx$ is the number of pixels per millimeter (CCD width)
  • Increasing cameraPixelNoise mask to 8x8 (and minUseGrad too) to correct the noise of the webcam and avoid cloud degeneration along the vision ray

4rd week

  • Adding clarifications in the technical documentation according to the suggestions of Javier Finat (thanks!)
  • Changed calibration configuration to ATAN camera model results in a point cloud with better borders but the depth computation is still buggy. Sometimes the tracker fails to recover the previous camera position without closing the loop.

phone-pointcloud

  • Correcting OpenCV practices written un Python of my students. This is the first time I can evaluate someone's work!

References