<<Prev 1 2 3 4 5 6 7 8 9 10 Next>>
Blob Filtering
From the previous slide we can see that the blue ball is very round while the detected camera
box is not. By adding in the Blob Filter module and adding in
a circular criteria above 0.65 we can eliminate all but the most circular blobs.
You will notice that 3 blobs have been detected. The 3rd blob is within the camera box and is a small
circular blob not connected to the rest of the box. This often happens as objects with pictures
will not always connect. To eliminate this false blob we could use the morphological Fill
module to fill in any holes within objects. This would cause the inner blob to merge with the rest of
the box and therefore become one object. We could also use the Convex Hull
to create a solid object with the same convex hull as the detected blob. While these techniques are certainly possible
we chose to add in another criteria within
the current blob filter to ensure that all detected objects are 'outer' objects. This filter criteria specifies
that if a blob is contained within another blob it should be eliminated from the current active blobs.
Adding this criteria yields:
To ensure that we get the closest ball we also add in a 'bottom' criteria to the blob filter. This will
ensure that a ball lower down on the screen will be given higher weights. Likewise, we also add in a
'largest' criteria to ensure that we go for the largest blue blob (this assumes that larger is closer).
Finally we also add in a 'belowY 120' criteria to ignore anything above the ground plane in case the
robot starts to track some guy in a blue shirt! Given all these criteria we begin to narrow the focus
of the robot to tracking blue balls that are within a certain size on the floor.
Next we have to go and grab the ball ...
<<Prev 1 2 3 4 5 6 7 8 9 10 Next>>
|
