Rtabmap real-time problem on Raspberry Pi2

[abilbrahem]

Hi guys :), I had successfully run Rtabmap on the the Raspberry Pi2 but I got just 3-5 FPS, so my question is can that be optimized by reducing point cloud density or range ? or over clocking the RPi2 ? or running ubuntu headless and just using an x server ? thank you so much.

[matlabbe]

Hi,

Not sure if it is what you are observing, but the FPS number shown at the bottom of the 3D Map view is the rendering framerate, not the actual odometry or mapping rate. The processing times for odometry and map update are shown under the Statistics view (Odometry/TimeEstimation and Timing/Total respectively).

For visualization, without a GPU the rendering would be slow. You can change how many points are shown per frame under the 3D Rendering panel in Preferences. You could disable odometry cloud rendering and just keep the map, then increase decimation parameter to reduce the number of points (you can also limit the depth).


To decrease odometry processing time, you could increase source decimation. This will reduce the size of the images. This may affect the odometry quality and loop closure detection though.


Another option is using ROS to split the core and visualization among two computers. With the ros package (rtabmap_ros), you can start the core (rgbd_odometry and rtabmap) on RPi2 and the GUI (rtabmapviz or rviz) on another computer.

cheers

[abilbrahem]

Hi Matlabbe, thank you again for the reply. about splitting the process in two how can that be done ? especially the part of connecting the pc and the RPi2 ? thank you.

[matlabbe]

Hi,

Do you know how to use ROS (tutorials)? There is a simple communication example on ROS here: http://wiki.ros.org/ROS/Tutorials/MultipleMachines

For example, you can set the ROS master on the RPi with rtabmap and rgbd_odometry nodes running (see example here), then start rtabmapviz on the remote computer by connecting rtabmap_ros/MapData topic.

For info, I tested the standalone yesterday on RPi3, some results here.

I'll try the ROS setup descrivbed above on RPi soon (EDIT see this post),
cheers