59
LOOP CLOSURE USING CONFIDENCE VALUES
Occasionally Stencil generates laser scan matches that are less than optimum. There are two ways
to fix this. You can change the parameters to reject these matches during raw data playback which
gives more flexibility (described later in this manual) or you can try to realign these scans during
loop closure. Confidence values are generated during scanning or raw data playback and recorded
in the trajectory file. The loop closure tool lets you define either a percentile as low as 0.1% or an
absolute value below which Stencil will stop adding scan frames to the current pose stack (if the
number of poses already added is higher than the minimum and lower than or equal to the
maximum) and attempt to match the pose stack back to the registered map. This method is useful
in realigning frame(s) for smaller adjustments, but is less useful in fixing significant drift or twists
in the trajectory due to bad visual odometry estimates due to occlusions or large moving objects
in the field of view. These are best fixed in the replaying of raw data by adjusting the parameters,
as well as confidence thresholds, and the visual odometry rejection parameters to generate better
results. The replay of raw data also allows better registration since the real time result is time
limited, whereas the Adaptive Data Replay Tool takes as much time as needed.
RUNNING LOOP CLOSURE TWICE
There are reasons you may want to run loop closure twice. You could choose to run loop closure
with GNSS data without loops enabled for a coarse correction of the trajectory file and then run it
with or without GNSS to do a much finer correction of the data. You would have a much higher
probability of finding the corresponding trajectory points that you are trying to line up after the
coarse correction was done. Otherwise the points might be too far away from each other to find
the corresponding points.
To Next Page
Loading...
