Also wondering if open3d development is going to be impacted (it was starting to be mature)
Awesome, we should talk !
Thanks! Feel free to shoot me an email at brandon at luxonis dot com.
Indeed, love to hear with Intel has in mind for Open3D: will it die or will it broaden to support an Intel CPU use case for any stereoscopic and ToF camera?
We just built our industrial reconstruction around Open3D. Intel Labs is technically a different group, but you could see how they would be impacted.
RealSense stereo cameras are mostly sticking around
RealSense CTO posted a comment in:
The author in this article has his doubts that the stereo cameras will be supported like before even though they will still be available for some unknown duration. Probably do not want to integrate it into a new product at this point I would guess:
Intel Will Keep Selling RealSense Stereo Cameras
Ah, you beat me to it. It’s close enough to being an official statement from Intel.
One thing though, doesn’t Autonomous Driving use Cameras and Computer Vision? I kinda thought that was Musk and Tesla’s thing. It would appear Intel is missing the boat on something again.
Okay, so they are keeping:
D410, 415, 430, 450, modules
D415, 435, 435i integrated product lines
Open Source SDK
Facial Authentication (F450) and Tracking (T265) product lines this month
D455 will be EOL’ed (but not the module)
A new version of the D455 may come later.
Oh phew … the L515 is actually kinda neat, a bit of a pity, but at least keeping the stereo cameras is a good thing since literally everyone uses them. Couldn’t care about the tracking line personally.
The closest that came to RealSense IMO was the Structure Core, many others including ZED and Mynt wanted to eat into my GPU and that wasn’t cool.
It’s kind of sad though they are EOLing D455. It seems like they picked the wrong product out of the hat to EOL?
The D435 has different field of view for RGB and depth, which IMO is the biggest design blunder in their entire line of products. You literally have to throw away half your depth data to create RGBD images. D455 fixed that and has similar FOV for RGB and depth. But they’re EOLing the fixed product and keep the buggy product … did the EOL person get their numbers mixed up?
I’m guessing companies like FRAMOS are still going to sell a productized version of the D450 module along with their other offerings.
But D455 has min distance 52 cm while D435 has 28 cm which is why I prefer the D435.
Same here for basic obstacle avoidance
ZED Stereo Camera is probably one of the better alternatives.
Espros Photonics www.espros.com has 3D TOF modules that are superior outdoor environments. TOFcam635 and TOFcam660 units are in full production and available on Digi-key
@ Uly I looked into it, it seems the EPC660 is $4752 on digikey.com? … might be good for some use cases but it seems to not be a good replacemenent for RealSense which is $189 - $279 … also for that money one could buy a couple of VLP-16 …
Unless they’ve changed anything it seems the ZED requires you to have GPU. The problem I had with that is for most robotics applications that use embedded GPU like Nano / Xavier / etc. that doesn’t leave enough GPU for neural networks. You also can’t exactly use it with a Pi, NUC, or anything without a GPU. A good replacement for RealSense should do all depth processing with an on-board ASIC.
The 635 is only $165 and looks interesting to me for two reasons: the min range of 0.1m and the dual beam. The resolution, 160 x 60, is terrible compared to the D435, but I’d be curious how that looks in practice. If the camera still detect objects much smaller than its resolution then it could still be a good fit for close-up obstacle avoidance.