What is the meaning of Realsense depth data?Distance from pixel to the center of the camera or distance of a depth plane?

[Alex_X_]

I used to use kinect v2,it's depth data means the depth of a vertical plane which the pixel belong to.

Dose Realsense work the same way?

[kfind1 · Accepted solution]

From my testing so far, a raw depth value is the distance from the point in the scene to the pixel on the image sensor (or equivalent sensing model, when using stereo vision) . I don't think the raw values are perpendicular to the X-Y plane, only after calibration is that true. Until you calibrate, you have to deal with the camera perspective/projection.

On the R200 I then get the device's calibration info and use a StreamCalibration and the Projection class to convert this raw depth to a proper world coordinate XYZ in floating point millimeter units values. This is still with reference to the camera frame - if you wanted a static world reference you would need to find/make the transform between camera and world frame, especially if the camera is moving.

[jb455]

You get the distance form the camera plane, i.e., perpendicular distance, not a straight line from the object to the camera lens.(is that what you meant?)

So if you had a completely flat object exactly perpendicular to the camera, the depth value should be constant across the face of the object (though in reality it's a bit wobbly due to errors)

[samontab]

Depth returned is the Z component of the 3D point with origin in the depth camera's optical center.

In other words, distance to the XY plane, not distance to the center of the camera.

[Alex_X_]

Thank you all, it helps a lot.