This step-by-step procedure shall guide you to identify good stereo matching parameters for your respective scene.
Capture Settings
Check the raw camera images: Is the images brightness suitable so that the projected pattern shows contrast on the targeted surfaces?
If your scene is fairly constant you might consider disabling auto exposure and auto gain and use fixed settings. Otherwise the automatic exposure and analog gain factor settings can be used for varying conditions. Adjusting Auto Brightness Target further allows you to adapt the image brightness when using automatic exposure, automatic gain, or both.
Note: Very long exposure settings may lead to variable illumination or trigger delays. Please also read the illumination topic for your camera model.
Stereo Matching Volume
First the depth range needs to be adjusted: Place your hand or another object at the required distance from the camera and adjust the sliders Maximum Distance and Depth Range so that the object is shown in the Depth Image.
Adjust 'Maximum Distance' and 'Depth Range'
Advanced Matching Parameters and Post Processing
Depending on the scene, the reconstruction with the default parameters may be incomplete or noisy. To maximize correctly reconstructed depth information you can adjust the parameters in the following systematic fashion:
1.Move all sliders to the left.
Start with lowest 'Change Cost' and 'Step Cost' parameters and no post processing
2.Start by increasing Depth Change Cost until most of the small-scale noise in compact surfaces disappears. Note that the Depth Step Cost must always be greater than Depth Change Cost and will therefore also increase automatically.
3.Increase the Depth Step Cost until also smaller low-contrast areas are correctly reconstructed while still no important surface details are getting lost.
4.Enable shadow filtering and check whether setting Shadowing Threshold to 0 or 1 has a positive effect on the borders of the interesting surfaces. A threshold of 1 will be more permissive and will leave more reconstructed pixels at object boundaries and steep surfaces. A threshold of 0 is more restrictive and will filter out more points, but it might result in more accurate estimates when using/computing surface areas of objects.
5.Increase Uniqueness Ratio until only small patches remain in the background
6.Increase Speckle Region Size to remove these small patches
7.(optional) Also increase Speckle Component Threshold to be more permissive and remove less regions from in bumpy or noisy surfaces.
8.Check and decide whether increasing Median Filter Radius to 1 or 2 has a desired effect on the smoothness of the reconstruction.
9.Increase Fill Region Size to close small gaps, stop before unwanted interpolation occurs.
10. (optional) Increase Fill Border Spread to make filling more permissive and also fill gaps whose surrounding surfaces are elongated in depth. Stop before gaps around object boundaries get filled.