I want to implement 3D pose estimation using Intel Realsense (D435). Is there any SDK available which can be used?

[MartyG]

Although the D435 does not have built-in support for pose estimation like the T265 Tracking Camera does, a couple of RealSense users have found workarounds for the D435 for use in applications such as picking items from a stock bin with a robot arm.

One of these users' approaches for getting pose with a D435 was an OpenCV SolvePnP algorithm.

https://github.com/IntelRealSense/librealsense/issues/3569#issuecomment-475896784

Another user offered a different approach involving ROS inference.

https://github.com/NVlabs/Deep_Object_Pose/issues/10

There is also an interesting related discussion about doing hand-eye calibration of a robot arm with D435.

https://github.com/IntelRealSense/librealsense/issues/3569

[RKuma119]

Thank you for this.It is beneficial. Does it have any built-in support for face detection? I am using my own DL algorithm for face detection, which is very much computational expensive.

[MartyG]

A Dlib face landmark detection example program was added in the new 2.30.0 version of the RealSense ​SDK.

https://github.com/IntelRealSense/librealsense/pull/4966

There are other face detection methods using OpenCV but they tend to be processing expensive too. A way to get a boost through hardware acceleration is to plug an Intel Neural Compute Stick 2 into an available USB port alongside your D435.

https://store.intelrealsense.com/buy-intel-neural-compute-stick-2.html