TY - GEN
T1 - On scale initialization in non-overlapping multi-perspective visual odometry
AU - Wang, Yifu
AU - Kneip, Laurent
N1 - Publisher Copyright:
© 2017, Springer International Publishing AG.
PY - 2017
Y1 - 2017
N2 - Multi-perspective camera systems pointing into all directions represent an increasingly interesting solution for visual localization and mapping. They combine the benefits of omni-directional measurements with a sufficient baseline for producing measurements in metric scale. However, the observability of metric scale suffers from degenerate cases if the cameras do not share any overlap in their field of view. This problem is of particular importance in many relevant practical applications, and it impacts most heavily on the difficulty of bootstrapping the structure-from-motion process. The present paper introduces a complete real-time pipeline for visual odometry with non-overlapping, multi-perspective camera systems, and in particular presents a solution to the scale initialization problem. We evaluate our method on both simulated and real data, thus proving robust initialization capacity as well as best-in-class performance regarding the overall motion estimation accuracy.
AB - Multi-perspective camera systems pointing into all directions represent an increasingly interesting solution for visual localization and mapping. They combine the benefits of omni-directional measurements with a sufficient baseline for producing measurements in metric scale. However, the observability of metric scale suffers from degenerate cases if the cameras do not share any overlap in their field of view. This problem is of particular importance in many relevant practical applications, and it impacts most heavily on the difficulty of bootstrapping the structure-from-motion process. The present paper introduces a complete real-time pipeline for visual odometry with non-overlapping, multi-perspective camera systems, and in particular presents a solution to the scale initialization problem. We evaluate our method on both simulated and real data, thus proving robust initialization capacity as well as best-in-class performance regarding the overall motion estimation accuracy.
UR - http://www.scopus.com/inward/record.url?scp=85031827635&partnerID=8YFLogxK
U2 - 10.1007/978-3-319-68345-4_13
DO - 10.1007/978-3-319-68345-4_13
M3 - Conference contribution
SN - 9783319683447
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 144
EP - 157
BT - Computer Vision Systems - 11th International Conference, ICVS 2017, Revised Selected Papers
A2 - Vincze, Markus
A2 - Chen, Haoyao
A2 - Liu, Ming
PB - Springer Verlag
T2 - 11th International Conference on Computer Vision Systems, ICVS 2017
Y2 - 10 July 2017 through 13 July 2017
ER -