Re: Incomplete depth cloud and no loop closures with Realsense R200

Posted by matlabbe on
URL: http://official-rtab-map-forum.206.s1.nabble.com/Incomplete-depth-cloud-and-no-loop-closures-with-Realsense-R200-tp2934p2935.html

Hi Daniel,

1. The realsense camera is a stereo camera in IR. On some surfaces, it can give various results. You can open rqt_image_view to see the raw depth image from the sensor. If you can see the checkboard pattern in the raw images, the problem is not from rtabmap. On the second image, it looks to me that the bottom of the image is the table, which cannot be detected by the depth sensor. The right black border is an effect of the registration of Ir to RGB

2. Loop closures can only been found on images with a lot of discriminative visual features. A white wall like this will give nothing. Revisiting locations should be also seen with the same view point. If the camera is seeing an empty space and has limited range, this would give empty depth images, which will affect loop closure performance (depth values are required). All these cases are similar to events that would cause lost odometry.

Visual odometry is not required for loop closure detection, rtabmap and rgbd_odometry nodes are independent. If you have already wheel odometry + IMU fusion, you can remove rgbd_odometry node to save a lot of computing resources.

To save computing power for rtabmap node, we can reduce rgb image size so that feature extraction is faster. To do so, similar to Tango ROS Streamer example, set:

<node pkg="rtabmap_ros" type="rtabmap" name="rtabmap">
   ...
   <param name="Mem/ImagePreDecimation" value="2"/>  <!-- use 2 times smaller RGB image to extract features -->
   <param name="Mem/ImagePostDecimation" value="2"/> <!-- save in database 2 times smaller raw images -->
</node>
</node>

cheers,
Mathieu