Keeping Maps Fresh: The Temporal Challenge

The world changes. New buildings appear, old ones are demolished. Stores rebrand, signs get updated. Seasons cycle. Our maps must keep up.

The Staleness Problem

Map captured: January 2021 User queries: December 2021

In between:

• Construction changed building facade
• Seasonal foliage blocked features
• Street furniture was moved
• Business signs updated

If >30% of features changed, localization may fail.

Change Detection

Query-Time Detection

When user query doesn't match map well:

• Low inlier ratio → something changed
• Inconsistent matches → partial changes
• High reprojection error → systematic shift

Can detect but not fix during query.

Proactive Monitoring

Periodic re-capture or user contribution:

• Schedule re-capture for high-value areas
• Collect user images (with consent) as passive monitoring
• Compare new captures to existing map

Automated Analysis

Detect changes without ground truth:

• Feature stability: Which features persist across time?
• Semantic consistency: Building still looks like building?
• Geometric consistency: Structure matches expected?

Handling Changes

Graceful Degradation

When map is stale:

• Return lower confidence score
• Fall back to GPS if confidence too low
• Suggest user contribute updated imagery

Multi-Temporal Maps

Store multiple versions:

• Seasonal variants (winter vs summer)
• Time-of-day variants (day vs night)
• Historical versions for change tracking

Query matches against appropriate variant.

Continuous Refinement

Update maps incrementally:

• New observations refine existing structure
• Change detection triggers re-mapping
• Automated quality checks before publishing

Economic Considerations

Re-mapping isn't free:

• Capture cost (vehicles, people, storage)
• Processing cost (compute for SfM/MVS)
• Validation cost (QA, review)

Prioritization strategy:

• High-traffic areas updated frequently
• Popular AR experience locations prioritized
• User-reported staleness triggers review
• Budget allocation by expected value

The Seasonal Challenge

Seasonal changes are predictable but dramatic:

• Deciduous trees: Full foliage → bare branches
• Snow cover: Completely changes appearance
• Sun angle: Different shadows throughout year

Options:

• Seasonal map variants (expensive, multiplies storage)
• Season-invariant features (research direction)
• Semantic matching (match building, not texture)

Currently exploring season-invariant feature learning.

Success Metrics

Tracking freshness:

• Map age distribution (how old are our maps?)
• Change rate by area (how fast do different regions change?)
• Localization success vs map age (correlation?)
• User-reported staleness incidents

Target: under 5% of queries affected by stale maps.