Skeleton Tracking SDK for D455 or D435 Inquiry
Hi! I'm a novice in this space, and I am in need of a bit of help.
Since I am particularly interested in using the camera and software to develop in the field of hand tracking and gesture recognition, I was wondering if the D455 or D435 depth camera would be a good choice in using the Skeleton Tracking SDK for my field of concentration. I also have a few related questions:
- Does using the higher-end depth camera necessarily help in producing faster and more accurate results especially in real time hand tracking and gesture recognition?
- Is the Skeleton Tracking SDK capable/efficient in hand tracking and gesture recognition applications?
- Would the Skeleton Tracking SDK work without a VPU (e.g., Intel® Movidius™ Neural Compute Stick 2)?
- And is the general process of developing and using the Intel depth cameras and software taxing on computer hardware? (as I am only using an 8th gen i7 ultrabook with 8gb of RAM, and I was worried that it would not be enough for this use case)
Thank you so much for your help in advance!
-
Hi Avino1033 The new RealSense D455 has a fast "global shutter" on its RGB sensor for the first time to match the fast global shutter on the depth. So if your project will use RGB color, then the fast RGB shutter of the D455 should help to avoid blurring on the color image during quick hand motion that the earlier 400 Series models were vulnerable to.
The Cubemos Skeleton Tracking SDK does not require a VPU such as Movidius.
The 400 Series cameras can work with any computer / computing device with an Intel or ARM processor. A saying that I like to attribute to it is that "it can run on a potato". This is possible because the camera has a built-in processing board called the Vision Processor D4 that can do processing work on the captured camera data that a computer's GPU would have to otherwise do if the D4 was not in the camera.
A single 400 Series camera can work on a low-power computing device such as a Raspberry Pi 3B. Having said that, RealSense applications can benefit from a good hardware specification on a computer. An 8th generation i7 should provide good performance with a single camera, as an i7 computer can handle four 400 Series cameras attached to the same computer.
It is also possible to offload work from the computer's CPU to the GPU to further accelerate an application. And the cameras can be optionally paired with the Neural Compute Stick 2 to benefit from hardware acceleration of vision computing projects.
My understanding is that for hand gesture recognition, the Nuitrack SDK commercial software may be a better option than Cubemos. Nuitrack SDK is compatible with the RealSense D415, D435 and D435i cameras. Cubemos supports "body gestures".
For advanced hand tracking, you may need to explore the advanced-level subject of training a camera with algorithms to recognise body joints. Intel have published a seminar on YouTube about doing this with RealSense.
Please sign in to leave a comment.
Comments
2 comments