Realtime PCL computation with Kinect

asked 2015-06-05 05:14:57 -0500

Arowana gravatar image

I am working on a realtime lightweight obstacle detection package. This package will help a mobile robot to sense obstacles. I use Kinect to get a point cloud of a scene.

My algorithm is running in 0.2s The conversion with pcl::fromROSMsg requires 0.060s The PassThrough filter requires 0.1 s And some other filters.

So the first two functions consumes a lot but looks quite simple to me.

Is there any way to downsample the Kinect pointcloud with cheaper function? By setting some freenect parameters for example... Or do you have any advice to help me mqke it faster?

I would like to go below 0.1s to have a good reactivity when the robot is moving

Thank you

edit retag flag offensive close merge delete

Comments

I found people talking about way to downsample the cloud. It looks to not easy at all. First way is to downsample the cloud in the callback, manipulating the point cloud is CPU expensive as it contains 5 millions points. An other way is to tweak the freenect/OpenNI driver to publish less points

Arowana gravatar imageArowana ( 2015-06-08 00:55:01 -0500 )edit
Arowana gravatar imageArowana ( 2015-06-08 00:55:12 -0500 )edit