| ROS Resources: ROS Homepage | Media and Trademarks | Documentation | ROS Index | How to Get Help | Q&A Help Site | Discussion Forum | Service Status |
Hi!
I’m wondering about investing in one Intel Realsense camera. Does anyone have an opinion on how good/bad they are? From what I’ve seen on youtube the sensors seemed quite noisy, on the other hand it looks like they might be good enough for basic tasks and the price is super attractive.
Would appreciate any insights you might have!
Mat
They are a little bit hard to install. The installation instructions aren’t great and you have to get one package from source. I find the point cloud to be noisy at a distance but very good at shorter range (like, less than 2 meters).
The ROS wrapper has a lot of functionality. I think it may be the best ROS camera package I’ve used. You can pause any camera at any time, which is cool if bandwidth is a concern.
The mechanical aspects of the camera are pretty good (robust cable connections, small package size, multiple mounting options). Personally I don’t think there’s a better option at the moment.
Thanks @AndyZe! Your points are very helpful! Do you think it’s worth paying ~30$ more for D435 to get a global shutter and a slightly larger field of view compared to D415?
I’m sure I’ll manage incomplete instructions (and maybe even create some docs while setting everything up).
To me, the wider field of view is definitely worth it. I don’t care about the global shutter, though.
I wasn’t thrilled with the accuracy of the D435. Emailed Intel and they said the 415 was more accurate (they replied right away which was a nice sign). The 415 still needs filtering but is adequate for me, the 435 was off by >2cm at distances less than 2M and very noisy. Their Windows app has a bunch of filters that were not implemented in ROS last I checked so I just wrote my own.
There was also an issue where sometimes you have to reset the device. Either soft reset or unplug and plug back in. This was an open issue on their github a while ago, not sure if its still there.
Otherwise I’ve been happy with it.
Thanks for your advice @AndyZe, @airduke, In the end I’ll go with 435 and will create some filters if needed to handle the noise. Can’t wait to start creating something!
As far as I know the issue that requires a reset is still there. I ran across a fork of the realsense2-camera package that actually detected this and did the soft reset automatically, but it’s still annoying that you don’t get frames for a moment. Apparently this is a firmware issue that is on their to-do list for some time already.
Some colleagues and me have mainly been using the D415 camera for a while now. Mainly the slow delivery of the D435 made us start with the D415 instead, but we’ve now received a D435 as well. I haven’t tried the D435 yet, so I can’t really say much about how they compare.
I’ve been playing around with a D415 in a mobile robot where I use it together with pointcloud_to_laserscan to generate laser scans for amcl and some other ideas for localisation. I was positively surprised by the accuracy of the depth image, where sometimes even up to 15 meters it gives reasonable measurements (less than 30 cm error). However I have only played around, no extensive checks and no tests in poor lighting conditions etc.
@Mbuijs thanks for your insights! Would you mind sharing what kind of computer did you use on the mobile robot? I’m wondering whether anything like Raspberry Pi 3 B+ can handle working with those cameras or should I rather look into Up Board or even Jetson TX2. I know it will mostly depend on what I do with the data but having some reference would be really useful.
I’m using my laptop currently, it’s a Lenovo ThinkPad T540p with an i7. I have no experience using it with a raspberry pi, sorry about that.
One thing I forgot to mention that I don’t use the pointcloud from the realsense2-camera, instead I take de depth image and use depth_image_proc to convert it into a pointcloud. This results in bigger pointclouds and somewhat higher cpu loads.
(post withdrawn by author, will be automatically deleted in 24 hours unless flagged)