Re: Mapping performance and hallway problems with Kinect 2

Posted by matlabbe on
URL: http://official-rtab-map-forum.206.s1.nabble.com/Mapping-performance-and-hallway-problems-with-Kinect-2-tp3956p3957.html

Hi Wezza,

1) Unfortunately, environments like that cannot be mapped robustly using a single kinect without other kind of odometry. At our lab, we have the same problem, it is why we are using relatively accurate wheel odometry combined with a lidar to navigate in feature-less corridors for most of our experiments. If you can afford a lidar, I highly recommend it. If you can change the environment, adding posters on the wall would help to have more visual features to track. Note that I also tried Google Tango (which is a quite accurate visual inertial odometry approach) in such environements, and the results are a lot worst (1 to 2 meters drift when the fisheye sees only a white wall) than wheel odometry alone (even without a lidar refinement). Many of the visual-only SLAM limitations are discussed in the following paper where we used wheel odometry and only a kinect: Cheap or Robust? The practical realization of self-driving wheelchair technology.

2) You may want to increase the framerate of the published images. For visual odometry, at least 10 Hz is preferred, or even more if your wireless network can support as much as data. Beware that WiFi connection would not be as good everywhere in the building, in particular the robot has to switch between wifi spots while navigating. In that last case, having odometry and mapping running onboard would be preferred. Make sure your computers are synchronized, that could fix some difficulties to synchronize topics and TF on the desktop side.

cheers,
Mathieu