Closed liquidcronos closed 6 years ago
We also need to know how we handle regions of good points. Do we allow new feature detection at risk of duplicating the same feature or do we disallow them at risk of losing immobile space to draw features from.
We start by calculating n clusters of points and generate a bounding box or convex hull for them. We can construct a mask from those which describe the available pixels for feature extraction.
Solved with bounding box for moving features and circles around static features
If the feature count is below a certain threshold we want to generate new points. Those points should not be generated in the same spot as the old moving points or the still existing immobile ones. To achieve this we want to segment the image into areas of activity and inactivity using the Moving Features. Using those Regions we want to find the new features in regions which had no moving points.