Re: How does rtabmap deal with dynamic objects?

Posted by PeterGa on
URL: http://official-rtab-map-forum.206.s1.nabble.com/How-does-rtabmap-deal-with-dynamic-objects-tp4735p8153.html

Thanks.

Placing a wall behind the dynamic object helped the removal of "outdated" points from the octomap_grid. But it still takes several approaches with the drone from different angles until the outdated points from the dynamic obstacle are finally removed. Is this behavior expected when using only a rgbd camera or should the point removal work faster? Would a lidar sensor improve the clearance performance?

I've created a screen cast which shows that behavior. I hope the provided link works for you.

https://www.dropbox.com/s/ohp4l6yb6j8oa4u/Octomap_grid_with_wall.mp4?dl=0

On the video you will see the drone flying around in the simulated world. On the bottom left there is the rgb image of the drone's camera and the the depth image of that camera. Note that the depth image is run through an algorithm that tries to filter out dynamic object in the image (which only works to some degree). So only the the points of the dynamic object that make it through the filter will be shown in the octomap_grid.

Do you think that there is a possibility to further improve the clearing of the dynamic obstacles based on the video? For instance using additional sensors or a higher update frequency for the octomap_grid?

Thank you really much for your help.