Poor pointCloud reconstruction

[Axel C]

Hi guys !
Can anyone tell me how i can improve the 3D reconstruction of my pointcloud ? My robot is visiting pipes and i would like to obtain a proper 3D cloud of it. It seems like local maps aren't proprely assembled in the global map. Any ideas ?
The odometry is correct : i made a round trip as a benchmark for the rgbd_odometry. Is it something related with the camera frame and the robot frame ?





Two different pipes, yet it's the same issue.
I believe the problem is related with

 But I don't know how to solve it..
Thanks for your Help ;)

[matlabbe]

Hi,

Looks like there is a missing optical rotation transform for the camera topics (yaw/pitch/roll: -pi/2 0 pi/2). Update your URDF with this transformation, then make sure the camera topics are using their respective optical frame.

cheers,
Mathieu

[Axel C]

Hi Matthiew,
I believe my URDF is correct as my realsense are set verticaly.
But I've set my RTABmaps parameters so that :

And I am no quite sure whether i have to set Rtabmap's frame_id to base_link or camera_realsense_1_link. The odometry is in the base_link frame but pictures are in the camera_realsense_1 frame and my tf_tree looks like this :


I am quite struggling with those transforms issue since i'm quite unfamiliar with ROS.
Thanks for the former response btw !
Axel

[matlabbe]

Hi,

If you do:

rosrun tf tf_echo base_link camera_realsense_1_link
rosrun tf tf_echo camera_realsense_1_link camera_realsense_1_color_optical_frame

and

rostopic echo /camera_realsense1/color/image_raw/header 
(well the name of the color image of your realasense1!)

What is it shown?