Outlier Removal#

The Outlier Removal feature available on blaze cameras allows you to remove pixels that differ significantly from their local environment.

Using the Feature#

How It Works#

An internal algorithm determines whether a pixel's depth (measured in gray values of the depth map) differs too much from the depth of its neighboring pixels. Pixels identified as differing too much are called outliers and will be classed as invalid. By enabling the OutlierRemoval parameter, these outliers will be removed from the output.

With the OutlierRemovalThreshold parameter, you can specify the strength of the filter, i.e., the number of inliers (good data values) in a 3x3 neighborhood required to make a pixel an inlier und therefore to prerserve it. The higher the value, the more outliers will be removed.

The OutlierRemovalTolerance parameter allows you to specify the tolerance of the filter. The lower the value, the more outliers are removed. With this parameter, you can specify the maximum permissible relative distance for a pixel to be classed as an inlier. Relative distance means the distance of a pixel to the center pixel. The lower the value, the more outliers are removed.

The image below shows invalid data, i.e., outliers, in an image. In this example, the outliers create a tapering effect on the far side of the object. In other images, the effects may be different.

The following image shows an image where the outliers have been removed.

Representation of Outliers#

The representation of outliers varies by component:

Configuring Outlier Removal#

1. Set the OutlierRemoval parameter to true.

2. Set the OutlierRemovalThreshold parameter to the desired strength.
   

   Info

   In most cases, best results are obtained with values between 3 and 7.

3. Set the OutlierRemovalTolerance parameter to the desired tolerance.
   

   Info

   • Start with a higher value (e.g., 50) and, if necessary, reduce the value until outliers are suppressed sufficiently.
   • Don't reduce the value any further to avoid removal of valid data.