Nice project! Your 3D printed case for the stereo pair removes a lot of potentially useful information for the feature extraction algorithm. If it can help, I have an old project where we mounted a realsense (R300 if I remember well) on Hololens.

You can inspire from our 3D printed mount, the CAD is available here : http://vision.gel.ulaval.ca/~jflalonde/projects/hololens3d/i...

This looks great! In reality, our casing only covers really small area, but we are always looking to improve :).

In case you're wondering - it's $209 from Mouser: https://www.mouser.com/new/Intel/intel-realsense-camera-t265...

It's also available for $199 right from Intel's site[0].

[0] https://www.intelrealsense.com/tracking-camera-t265/

Interesting. I wonder how it compares in capability to these algorithms, which are in Nvidia’s Redtail SDK:

https://youtu.be/T7-IxCW1UlA

Neat, thanks for linking this, I hadn't heard about Redtail before.

This camera only tracks its own pose - it doesn't output a full depth map, according to the FAQ.[1]

The visual odometry algorithm ("Elbrus") in that video can supposedly be run on an NVIDIA Jetson board (unclear which one) at 30fps.[2] It seems like this runs on the CPU, which is pretty impressive.

The Elbrus demo app uses the $450 ZED stereo camera[3], compared to the $200 T265[4], though there's probably nothing stopping someone from using Elbrus with cheaper cameras, including possibly this one.

  1. https://www.intelrealsense.com/tracking-camera-t265/#fl-accordion-5c70b553f2112-tab-0, "Is this a depth camera?"
  2. https://docs.nvidia.com/isaac/isaac/packages/perception/doc/visual_odometry.html
  3. https://store.stereolabs.com/products/zed/
  4. https://store.intelrealsense.com/buy-intel-realsense-tracking-camera-t265.html

I have a Jetson Xavier and dual $50 PS4 stereo cameras on my robot and will be trying the Nvidia algorithms.

https://youtu.be/crHUGLr-JMI

I bought an Intel Realsense last year (2 generations ago) and this is much more powerful, esp. in localization . If you buy this, expect it to be obsolete in about 6 months

Is there a "hobby scale" way to do SLAM?

Why does Intel make a product like this ?

Those cameras were developed by an Israeli startup that Intel bought a few years ago (Omek Interactive). Making these sensor low power could enable devices to have 3D reconstruction and tracking. This opens the door for more robust AR/VR applications and cheap devices for robotics. Not only that Intel is in a good position to sell/produce those sensors, but they probably sense that it could be a good market to conquer.

That's half of it. The other half is that Intel has pushed the RealSense API as something that is independent of the exact technology used.

So this could be something that uses a projected dot matrix like the Xbox 360 Kinect or a time-of-flight sensor like the Xbox One Kinect or multiple conventional cameras or...

There are many neat things you could do with depth cameras in principle, in particular make computers into "sessile robots" that appear to understand the space we live in the same way we do. Unfortunately the market hasn't made this into reality yet.

Hoping that it's one of those things that will soon be in every phone. Like modems.

Some of the older RealSense depth cameras are already integrated in laptops and tablets

Does that look like micro-USB? Like, is it really 2019 new Intel tech being released with micro USB and not USB-C?

This reminded me of the Microsoft Kinect with the 2 cameras for tracking

Back up a bit... what exactly is a tracking camera?

It's a 'Simultaneous Localization and Mapping device' (SLAM) that will give you its position, velocity and acceleration in 3D. There's quite a large [1] literature on the subject.

[1] http://www.arxiv-sanity.com/search?q=slam