Intel RealSense

SUMMARY

The RealSense class provides a Python interface for Intel RealSense depth cameras, supporting single-frame capture of RGB images, depth images, and 3D point clouds, as well as continuous video streaming in all three modes. It wraps the Intel RealSense SDK (librealsense) to handle device connection, stream configuration, and safe teardown.

Requirements: The Intel RealSense SDK (librealsense) must be installed and a supported RealSense camera must be connected via USB.

Tested devices:

• RealSense Depth Camera D405
• RealSense Depth Camera D435if

Useful Links

• RealSense Quickstart
• Official SDK

Installation

Install Medulla first — see Installation.

Then install the RealSense-specific dependencies. It is possible to either use a precompiled SDK or build it from source. In the following we will only consider the precompiled SDK.

Download and Install Precompiled SDK

1. Download the latest SDK bundle from the Releases page.
   • See also: Windows distribution guide
2. Connect your RealSense camera.
3. Run the included tools to test the installation:
   • RealSense Viewer: View streams, tune settings, record and playback.
     • Install the firmware for the connected camera from the RealSense Viewer.
     • Start the depth stream and RGB stream by toggling the "Stereo Module" and "RGB Cameras" to "on".
   • Depth Quality Tool: Measure accuracy and fill rate.

RealSense Viewer after successfully installing and connecting the camera.

Python Packages

Install the RealSense dependencies via the Telekinesis SDK:

pip install telekinesis-ai[medulla-realsense]

Getting Started

WARNING

Examples were tested on Windows 11 only.

After testing the installation with RealSense Viewer, make sure to either close the viewer or switch off the streaming modules on the left side. Then you can capture a pointcloud with a RealSense camera from medulla with:

from telekinesis.medulla.cameras import realsense

cam = realsense.RealSense(
    name="my_realsense_camera"
)
cam.connect()
pc = cam.capture_pointcloud()
print(pc.positions)
cam.disconnect()

Examples

Skill	Description
Capture Color Image	Capture one colour image and visualise with Rerun
Capture Depth Image	Capture one depth frame with colour map and visualise with Rerun
Capture Point Cloud	Capture one colour-mapped 3D point cloud and visualise with Rerun
Stream Live Color Video	Stream continuous colour frames to a Rerun viewer
Stream Live Depth Video	Stream continuous depth frames with colour map to a Rerun viewer
Stream Point Cloud	Stream a colour-mapped 3D point cloud continuously to Rerun
Get Intrinsics	Retrieve camera intrinsic parameters for calibration workflows