Wrong Loop Closure in Outdoor Effect the Map

[imran05]

Hello,

I am using RTAB-Map for mapping an orchard environment. I’ve used RTAB-Map before for indoor and outdoor mapping, and it worked well. However, this time I ran an experiment in a field with artificial trees, and I’m facing problems.

RTAB-Map is detecting wrong loop closures because most of the detected features are from the background. I tried setting RGBD/OptimizeMaxError=0.1, which rejected the wrong loop closures, but the trajectory is still poor, and the map quality is bad.

Can anyone suggest which parameters I should adjust to improve the results? I am attaching the rosbag recorded during the experiment and the database for reference.
Rosbag: https://drive.google.com/file/d/1okPECevkLlV2g4kLIHLr_oDhx4pewcLF/view?usp=drive_link
Database: https://drive.google.com/file/d/18jIt8DMWybzQv1qtFAbJi13pDwYNY2Jx/view?usp=sharing

[matlabbe]

The database doesn't contain any loop closures, what you are seeing is probably visual odometry drift. I could not test "as is" your bag as the camera_info is not what rtabmap expects:

[ERROR] [1737327595.911208693, 1737093742.310247645]: The stereo baseline (-0.074953) should be positive 
(baseline=-Tx/fx). We assume a horizontal left/right stereo setup where the Tx (or P(0,3)) is negative in the right camera info 
msg.

That could probably fixed if the camera driver publishes the left camera_info with a positive P(0,3) (currently negative).

About the odometry drift, I think there would be enough good visual features on the ground. However, the biggest issue is that the stereo camera is poorly calibrated, there is a noticeable vertical disparity:

[imran05]

That error is happening because the camera driver has a opposite camera positions in the depthairos driver, For removing that error i just replace thge left and right cameras. I am suing below command:

roslaunch rtabmap_launch rtabmap.launch \
    args:="--delete_db_on_start" \
    stereo:=true \
    left_image_topic:=/stereo_inertial_publisher/left/image_rect \
    right_image_topic:=/stereo_inertial_publisher/right/image_rect \
    left_camera_info_topic:=/stereo_inertial_publisher/right/camera_info \
    right_camera_info_topic:=/stereo_inertial_publisher/left/camera_info \
    imu_topic:=/stereo_inertial_publisher/imu/data \
    frame_id:=base_footprint \
    approx_sync:=true \
    approx_sync_max_interval:=0.001 \
    wait_imu_to_init:=true

For the camera, I calibrated it many times but i am still facing that problems.So, I will try to calibrate it agin.

Thank You

[matlabbe]

The depthai tool would calibrate with right camera as reference instead of left. That is fine, as long as it is reflected correctly in the camera_info generated. Maybe it is just a bug on how they publish the camera_info:

That could probably fixed if the camera driver publishes the left camera_info with a positive P(0,3) (currently negative).