Pointclouds from Azure Kinect rgbd depth image are distorted

[chz31]

Hi all,

I am doing some pilot studies about using the rgbd image taken by Azure Connect to extract pointcloud and reconstruct a surface model of the object as a potential cheap way to collect shape data from biological specimens. However, the point cloud of the object (a tea box) from the 3D view in the Azure Kinect viewer appears to be distorted as the edges are not straight as the screenshots shown (I focused the camera on the tea box as in the first picture). Further. one part of the turn table that holds the tea box at the left side (3rd screenshot) appears to be pushed back to the background.

I tried different resolutions and depth modes and moved the objects around, but the problem persists. Does anyone encounter a similar issue? Is it because of any hardware calibration issue or is there a way to adjust the configurations to get more accurate depth information?

Thank you very much!

Desktop (please complete the following information):

• OS with Version:
• SDK Version:
• Firmware version: <!-- RGB camera: 1.6.110
• Depth camera: 1.6.79` -->

[qm13]

Azure Kinect uses IR time-of-flight technology and you are seeing multipath distortion. Multipath distortion occurs when the IR reflection reaches the sensor via multiple paths. This is typically seen with angular surfaces e.g. point the camera at the corner of a room and you will see curving of the edges.

[chz31]

@qm13 Thank you for letting me know! I'll try it again on objects & backgrounds with smoother surface.