Geometry & SLAM
On Scale Initialization in Non-overlapping Multi-perspective Visual Odometry
International Conference on Computer Vision Systems · 2017
RESEARCH SUMMARY
This paper studies metric-scale initialization for visual odometry with non-overlapping multi-perspective cameras, including the motion degeneracies that can make scale unobservable.
Read the supporting paper ↗
THE RESEARCH QUESTION
Does a known multi-camera baseline always make motion scale observable?
A calibrated rig has metric geometry, yet certain motions still fail to constrain translation scale. A reliable odometry pipeline must recognize and address this during initialization. [Paper]
How the method works
- The paper analyzes scale initialization in a non-overlapping multi-perspective setup.
- Its solution is integrated into a real-time odometry pipeline and evaluated with simulated and real observations.
RESEARCH INTERPRETATION
What this contribution means
The work distinguishes possessing a physical baseline from obtaining informative measurements of scale. This observability issue is an important starting point for later vehicle-mounted surround-view odometry research.
This section explains the contribution in context. For experimental protocols, numerical results and comparisons, consult the original paper.
Scope and limitations
The solution concerns calibrated multi-perspective geometry. It should not be interpreted as recovering absolute scale from an arbitrary uncalibrated monocular video without additional information. [Paper]
Questions about this research
Can a calibrated rig still encounter scale ambiguity?
Yes. The paper studies degenerate motions under which the available observations do not sufficiently constrain metric scale.
Read the original abstract
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.
Original publication record ↗Sources, paper & code
Primary research sources support the summary above. Project resources provide demonstrations or implementations; Google Scholar is a bibliographic index.
- Primary sourcePublisher — original publication record ↗link.springer.com
- Primary sourceFull paper — author-hosted PDF ↗1fwang.github.io
- Metadata indexGoogle Scholar — bibliographic listing ↗scholar.google.com