Mobile Depth-Sensing
Hey everyone,
I’m working on a mobile robot / drone (or point-cloud / 3D-scan) prototype, and I just picked up an Intel RealSense D455 (82635DSD455) to handle depth sensing. I wanted to share what I’m planning and ask for advice, since I know some of you have experience with these.
Here’s the setup I’m envisioning: I’ll mount the D455 onto the robot chassis using its 1/4-20 tripod thread for a stable platform. The 95 mm baseline between its depth sensors gives much better accuracy at longer ranges, Intel claims less than 2% error around 4 m.
Because the D455 has a global-shutter RGB sensor and an IMU (inertial measurement unit), I can use it to not just detect obstacles but also track motion better when the robot moves. I’m planning to run it on my laptop or embedded system via USB-C (it supports USB-C 3.1 Gen1).
For software, I’ll be using the RealSense SDK 2.0, since the D455 is fully supported and it’s cross-platform (Linux, Windows, etc.).
Here are a few things I’m still testing or trying to figure out:
-
How reliable is the depth data when the robot is moving quickly? Will the IMU help stabilize the depth estimates?
-
What’s the real-world working range when mounted on a moving platform — how well does it detect obstacles at, say, 4–5 meters?
-
When used in outdoor or mixed-lighting environments, how much noise or depth “flicker” happens?
-
For calibration: should I try the on-chip self-calibration (RealSense SDK supports <15 seconds) to improve depth accuracy?
If other forum members have used the D455 (or similar D400 series) on robots or mobile platforms, I’d love to hear your feedback. What frame-rate and resolution did you run for your depth stream, how did you mount the camera for best performance, and did you notice any quirks with the IMU?
Thanks in advance, happy to share my build progress too once I get past the prototyping stage.
-
Hi Mercio Jonova Thanks very much for your questions.
1. RealSense 400 Series cameras equipped with fast Global shutters on their sensors, such as D455, can handle depth sensing of movement as fast as a car travelling at full speed. This is demonstrated in a YouTube video where a D435 is mounted to a moving car.
https://www.youtube.com/watch?v=OwJmCyAn3JQ
The IMU component is not involved in depth generation or stabilization.
2. The optimal depth sensing distance of D455 is up to 6 meters - in other words, depth accuracy is very good up until that distance before it starts noticeably drifting beyond that distance.
3. Noise / flickering may occur indoors if the camera is used in a location with overhead fluorescent type strip lighting. Configuring a setting called Power Line Frequency can help to compensate for this interference.
Outdoors, RealSense 400 Series cameras work very well in both bright-light and dim conditions. If you have a scene where there are both bright and dim areas simultaneously, a feature called High-Dynamic Range (HDR) can be configured.
https://dev.realsenseai.com/docs/high-dynamic-range-with-stereoscopic-depth-cameras
If the camera is pointed directly at the sun then the sensors can become saturated with light, disrupting the images. If you anticipate that the camera may regularly be pointed at the sun frequently, you could change to the D455f camera model that filters out sunlight or purchase the D455f's filter separately and fit it over the sensors on the outside of your current D455.
4. If you want to improve depth accuracy through calibration then the Tare tool should be used, as On-Chip improves depth quality.
Please sign in to leave a comment.
Comments
1 comment