The video shows the unboxing and installation of the Structure Sensor to the Jetson TK1 development kit. You can find out more about the Structure Sensor here: Structure.
So why is this device interesting? You attach it to an iPad or iPhone after all. The key is that the device is based on PrimeSense technology, which is what the original Kinect 360s were based on. (Microsoft has since moved to a different approach for depth sensing for the second generation Kinect 2).
The PrimeSense (recently acquired by Apple) method of depth sensing uses a structured light technique. A simplistic explanation is that an infrared laser (!) is shone through a patterned film, the pattern on the film being known. An infrared camera then views the scene, and by determining the deformation of the pattern calculates the depth of each point.
Inside the open source community, there is a project called OpenKinect which enables interfacing with a PrimeSense device and provide depth streams, infrared streams, color streams, sound, etc. With the appropriate cables, the Jetson is able to interface with the Structure Sensor.
As the video shows, the Structure Sensor is considerably smaller than the original Kinect. It’s easy to imagine this being attached to a Jetson platform (robot, tablet, DSLR, etc) and used for depth sensing. The Structure does not have a color camera or microphones in it, but for a lot of applications this is a good thing. The Structure Sensor is enclosed in a sturdy brushed aluminum housing.
Occipital is supporting the open source community, and accomodates access to different platforms as a result, by hosting interface software OpenNI2 on a Github Repository. They also sell a ‘Hacker cable’ which allows the Structure to be plugged into a regular USB port.
All good stuff, and enough to start having some fun!
6 Responses
Hello!
You mentioned we can use the Structure Sensor attached to a DSLR. Can we do this without a computer in cases I want to capture something outdoors?
Hi Uma,
Unfortunately the Structure Sensor does not have local storage, so it needs to be attached to a computer of some sort. The original intent of the Structure Sensor is to connect it to an iPad (which has the computer built in of course). People who use it with a DSLR usually have a laptop or equivalent in a backpack that they carry when recording. Thanks for reading!
Did you use a cross compiler for the ARM install?
https://github.com/occipital/openni2
Cross-Compiling for ARM on Linux
The following environment variables should be defined:
ARM_CXX=path-to-cross-compilation-g++
ARM_STAGING=path-to-cross-compilation-staging-dir
Then, run:
PLATFORM=Arm make
Thanks.
I compiled it on board the Jetson. Thanks for reading!
How does the quality of the depth data compare to the RealSense D435 you reviewed elsewhere? Is one “better” than the other, for the RACECAR/J application?
It depends on the lighting conditions. The Structure relies on an infrared projector, so it cannot be used outdoors, where as the D435 can. I don’t really do comparisons to say which is “better” because it is so dependent on where and how the device is being used.