Mapping from two PointCloud sources + One RGB camera

[srijal97]

Hello! I have a drone with two ToF depth sensors onboard, publishing both pointcloud and depth image. There is also an RGB camera onboard, and I have camera_info topics for all these sources.
First, I have combined the two pointcloud sources using:

    <!-- Node which combines up to four pointcloud sources into a single stream -->
    <node name="cloud_aggregator" pkg="rtabmap_util" type="rtabmap_point_cloud_aggregator">
        <remap from="cloud1" to="/tof_left_pc_fixed"/>
        <remap from="cloud2" to="/tof_right_pc_fixed"/>
        <remap from="combined_cloud" to="/tof_combined_pc"/>
        <param name="count" type="int" value="2"/>
        <param name="frame_id" type="string" value="hires"/>
        <param name="fixed_frame_id" type="string" value="odom"/>
        <param name="approx_sync" type="bool" value="true"/>
    </node>

Note that hires is the RGB camera frame ID, and ToF pointcloud data /tof_left_pc_fixed and /tof_right_pc_fixed are in left_tof and right_tof frames respectively. I figured that having the combined pointcloud output in hires would make combining it with color information from the RGB camera easier.

I then passed this point cloud to RTAB-Map with the following arguments:

    <!-- RTABMap -->
    <include file="$(find rtabmap_launch)/launch/rtabmap.launch">
        <arg name="stereo"                      value="false" />
        <arg name="depth"                       value="false" />

        <arg name="subscribe_scan_cloud"        value="true" />
        <arg name="subscribe_rgb"               value="true" />

        <arg name="scan_cloud_topic"            value="/tof_combined_pc" />
        <arg name="rgb_topic"                   value="/hires/image_rect_color" />
        <arg name="camera_info_topic"           value="/hires/camera_info" />

        <!-- <arg name="depth_topic"                 value="/tof_combined_depth/image" /> -->
        <arg name="frame_id"                    value="base_link" />
        <arg name="visual_odometry"             value="false" />   <!-- Launch rtabmap visual odometry node -->
        <arg name="odom_topic"                  value="/qvio/odometry_corrected" />
        <arg name="odom_frame_id"               value="odom" />
        <arg name="approx_rgbd_sync"            value="true" />
        <arg name="approx_sync"                 value="true" />
        <arg name="rtabmap_viz"                 value="false" />
        <arg name="rviz"                        value="true" />
        <arg name="args"                        value="--delete_db_on_start"/>
    </include>    

RTAB-Map creates an uncolored map, and I would now like to include color information from the RGB camera. The combined pointcloud source has a broader field of view than the RGB camera, so I am unsure how to approach this problem. Would I have to "crop" out the pointcloud somehow to make it see the same pixels as the RGB camera? Is there a way to partially color the map where RGB and pointcloud pixels overlap? I have also tried using rtabmap_utils/pointcloud_to_depthimage to convert my combined pointcloud to a depth image, but this does not work and instead throws the error. Here are the launch arguments:

    <!-- Node which reprojects a point cloud into a camera frame to create a depth image -->
    <node name="cloud_to_depth" pkg="rtabmap_util" type="pointcloud_to_depthimage">
        <remap from="cloud" to="/tof_combined_pc"/>
        <remap from="camera_info" to="/hires/camera_info"/>
        <remap from="image" to="/tof_combined_pc/image"/>
        <remap from="image_raw" to="/tof_combined_depth/image_raw"/>

        <param name="decimation" type="int" value="1"/>
        <param name="fill_holes_size" type="int" value="0"/>
        <param name="fill_holes_error" type="double" value="0.1"/>
        <param name="fill_iterations" type="int" value="1"/>

        <param name="fixed_frame_id" type="string" value="odom"/>
        <param name="approx" type="bool" value="true"/>
        <param name="wait_for_transform" type="double" value="0.2"/>
    </node>

Even though fixed_frame_id is set to odom, I still get this error (it should be looking for hires to odom but the error shows hires to ):

[ WARN] - /cloud_to_depth: [1692077220.677414292, 1691442532.882630568] Could not get transform from hires to  after 0.200000 seconds (for stamp=1691442532.544901)! Error=". canTransform returned after 0.201555 timeout was 0.2.".

I am not much experienced with image processing/handling so it probably does not make sense to convert this combined pointcloud to a depth image because its too wide? Given my sensors, I would love to get some suggestions on how RTAB-Map should be structured. If it helps, here is a youtube video showing the pointcloud map created using RTAB-Map and a view of my RGB camera:

[matlabbe]

Hi,

If you have a database with point clouds  and RGB frames, you export the map point cloud with color using the rtabmap-export tool:

rtabmap-export --scan --cam_projection rtabmap.db

Otherwise, your approach looks good, however, the warning "Could not get transform from hires to  after 0.200000 seconds " tells that one of the input topics doesn't have its "header.frame_id" set.

cheers,
Mathieu