Decimate function

[denzle]

Hi Mathieu,

I hope you're well. I've successfully managed to get both lidar and stereo-depth sensors working extremely well for my woodland mapping project. But i have noticed that with the S-D sensors that due to information overload the systems dramatically slows down to almost a stutter. From your previous help and from other posts i understand that decimation should help with live mapping resource use. I was hoping though that you may be able to offer more information regarding the function, how it works and if there are any best practices that you could advise on ie during mapping, post mapping, processing etc

Cheers
Dan

[matlabbe]

Hi Dan,

A "decimation" of 2 means that for an image of 640x480, we get an image of 320x240. A decimation of 4 would result in an image of 160x120. When rendering point clouds, we then project 160x120=19200 points instead of 640x480=307200. As a comparison, a Velodyne VLP16 would produce 300 000 per second, while an image at 640x480x30Hz would produce 9 216 000 per second. When decimating those images by 4, we would get instead 576 000 points per second, which is more in the range of common 3D LiDARs.

rtabmapviz 3D rendering parameters can use high decimation parameters to avoid having visualization lagging because there are too much points to render. However, when exporting offline, we can chose to regenerate the clouds at higher resolution.

cheers,
Mathieu

[denzle]

Thank you so much for that mathieu.