Questions about D415 accuracy
Hello, I am trying to understand the Realsense D415 Accuracy details before buying it.
For now, I have the next questions:
1)
According to D415 docs (https://www.intelrealsense.com/depth-camera-d415/)
it has Depth Accuracy: <2% at 2 meters
According to D435 docs (https://www.intelrealsense.com/depth-camera-d435/)
it also has Depth Accuracy <2% at 2 meters
At the same time, many sources say that D415 is more than twice more accurate than D435. So how does it work?
2)
"2% at 2 meters" - is it rarely possible measurement error or an error that I should expect at almost any point of a frame (like a permanent noise)?
3)
A specific example of question 2 (not far from my real requirements).
I am interested in scanning the wall at a distance of 1.5 meters via D415 (the wall is parallel to the sensor.).
2% at 1.5 meters gives me 3 centimeters distance scanning error.
A small cube with edges of 2x2x2 centimeters is glued to the wall (note, less than 3 cm error).
I need to know if D415 can show at least some recognizable depth trace of this cube.
I mean I don't need to recognize the shape or size of this cube, but I need to know that the object is there, detect that it exists.
Would it be possible with D415?
-
Hi Dzhus Thanks very much for your questions.
1. Depth measurement error starts at around zero at the camera and increases linearly over distance. This phenomenon is called RMS Error. On the D435 / D435i camera models, error increases at a rate that is approximately 2x that of D415, meaning that at around 1.5 meters D415 and D435 / D435i will have similar levels of error, with D415 having a little less. The greater the distance from the camera, the greater the difference in error will become.
Section 5. Understand theoretical limit of Intel's camera tuning guide linked to below provides further information about RMS Error and a graphical chart demonstrating the D415 and D435 /D435i error change over distance.
2. Whilst RMS Error is one of the factors that can generate error, there are also other environmental and lighting factors that can contribute to error too.
Examples are fluorescent lights such as ceiling strip lights, and aspects in the observed scene that represent a 'repetitive pattern' of repeating horizontal or vertical elements, such as a horizontal row of fence posts or a vertical stack of window blind stats. There are usually methods that can be employed to reduce particular types of error though.
3. A 2x2 cm cube may be difficult to isolate from the rest of the scene around it at 1.5 meters range. It may help if the cube has a textured surface instead of being plain and smooth, as a textured surface will be more easily analysable for depth information.
Please sign in to leave a comment.
Comments
1 comment