Graphics Reference
In-Depth Information
100
90
80
70
60
50
40
30
20
10
0
700
600
500
400
300
200
100
0
0
100 200 300 400 500 600 700 800 900 1000 1100
0
100 200 300 400 500 600 700 800 900 1000 1100
Time Step
Time Step
Fig. 2.9
Graph complexity over time for Seq1, with and without reduction. The two regions
bounded by
vertical dotted lines
are periods of revisitation, during which views are reobserved
rather than created. The reduced graph complexity remains constant unless new views are created.
Note the difference in vertical scale
Figure
2.8
overlays the trajectories computed using the heavily optimized, fully
complex graphs with those computed using reduced graphs. The qualitative similarity
of the results reflects the quantitative similarity of the errors to ground truth. The
deviation between the two traversals of the large loop in
occurs because the
robot traverses in opposite directions, so views are not reobserved.
Figure
2.9
shows the growth in number of graph nodes over time for
Seq3
, with
and without reduction. Reduction keeps the complexity linear with number of views
rather than time.
Seq1
2.10 Conclusion
This view-based monocular SLAM system minimizes the computation required for
vision-based processing and actively manages the complexity of the SLAM graph to
permit operation on constrained computational platforms. Our results show that the
complexity reduction methods significantly limit graph node and edge cardinality,
while only negligibly affecting localization accuracy. The system uses inexpensive
sensors, has low computational requirements, and high reliability, all of which are
ideal for low-cost robotic applications.
References
1. Carlevaris-Bianco N, Eustice RM (2013) Long-term simultaneous localization and mapping
with generic linear constraint node removal. In: 2013 IEEE/RSJ international conference on
intelligent robots and systems (IROS), IEEE, pp 1034-1041
