Depth estimation measurement error


I’m using the zed2i to estimate the distance between the camera and object but the estimation error seems higher then expected. I was hoping someone can help us and share an explanation on this.

The object is located 3.9 meters away from the left camera and the estimated distance extracted from the relevant pixel in the depth map is 3.6 and 3.7 meters using InitParameters.depth_mode ultra and neural respectively. Point cloud values are relativity close to the depth map values.Also, please take note that the confidence level of the corresponding pixel is 0 (“tupple(success, 0.0)”). Measurements were done indoors under good lighting conditions and the texture of the object isn’t specular (coarse printed 3d object).

These measurements yield an estimation error 8-5% for a relatively close range with respect to the 20 meters distance from the camera specification.
Is there a need to re-calibrate the camera parameters ? does this results make sense in terms of expected accuracy ?


Can you send us a svo file so that we can investigate ? You should not have this amount of error. You get 20cm of error, you should get at most 3 or 4cm.
The calibration can be the problem. Are you in a cold or warm environment ? Is the camera old ? Did the camera receive any shocks? All these things can have an impact on the calibration.
You can try a manual calibration with our tool ZED Calibration.