[CHI/CSCW Research] Virtual City Explorer: Scaling Urban Mapping via Remote Crowdsourcing
On the mapping of Points of Interest through StreetView imagery and paid crowdsourcing
The paper introduces the Virtual City Explorer (VCE), a crowdsourcing system that utilizes Street View imagery and paid workers to map Points of Interest (PoIs). By combining remote exploration with the DBSCAN clustering algorithm, it achieves high area coverage and provides a scalable, low-cost alternative to traditional on-site expert surveys and volunteer-based geographic information.
TL;DR
Mapping every bike rack or disabled parking spot in a city is a logistical nightmare. This paper presents Virtual City Explorer (VCE), a system that lets paid crowdworkers "walk" through cities virtually using Street View to identify Points of Interest (PoIs). By implementing a "Taboo" strategy that prevents workers from reporting the same item too many times, the researchers boosted discovery rates by 30% while keeping costs significantly lower than professional on-site audits.
Problem & Motivation: The VGI Bottleneck
Traditional mapping relies on two pillars: Experts (expensive, slow) and Volunteers (unpredictable, geographically biased). This is known as Volunteered Geographic Information (VGI). If you need to map a specific neighborhood in a week, you might find zero volunteers living there, and sending a city official could cost hundreds of dollars.
The authors' insight was to leverage the "Flat Earth" model: why should a mapper be physically present? With high-resolution Street View imagery, a worker in a different country can audit a street in London or DC just as effectively, provided they have the right tools.
Methodology: Triangulation and The Taboo Optimization
1. The Triangulation Interface
To ensure accuracy, workers aren't just clicking a point. They must "take a photo" of a PoI from three different angles. The system then calculates the intersection of these vectors to pinpoint the exact GPS coordinates.
Figure 1: High-level architecture of the Virtual City Explorer (VCE).
2. The DBSCAN Aggregator
Individual worker reports are noisy. The system uses DBSCAN (Density-Based Spatial Clustering of Applications with Noise) to group points within a 10-meter radius. If at least 3 workers identify the same spot, it is confirmed as a PoI.
3. Solving the "Easy Pickings" Problem (Taboo Strategy)
In early tests, workers flocked to the obvious PoIs near their starting positions, wasting budget on redundant data. The authors introduced Taboo PoIs: once an item is confirmed by 3 workers, it appears with a red "Taboo" sign in the virtual world. New workers are blocked from reporting it and must explore further afield to earn their reward.
Figure 2: The worker interface showing the Street View explorer and the "snapshot" reporting tool.
Experiments & Results: Efficiency at Scale
The researchers tested VCE in Trento (Italy) and Washington D.C., targeting bike racks.
- Cost-Effectiveness: In Trento, the VCE cost $14.40 to map 27 bike racks, compared to €60 for a government expert.
- The Power of Taboo: In the Trento study, the Taboo strategy led to a 30% increase in unique PoI discovery compared to the basic "free-for-all" method.
- Completeness: Interestingly, while VCE missed some PoIs found by experts (due to visual obstructions in Street View), it also found new PoIs that experts had missed, proving its value as a complementary tool.
Figure 3: Heatmaps showing worker paths. The Taboo strategy (right) forces workers to deviate from main paths into less explored segments.
Critical Analysis & Conclusion
VCE proves that virtual urban auditing is not just feasible but economically superior for specific tasks.
Takeaways:
- Redundancy is Waste: Without the Taboo mechanism, 88% of effort was spent re-confirming already known points.
- The "Escape" Condition: To keep the task fair, the authors added an "escape" clause—if a worker walks a long distance without finding anything new (because the area is fully mapped), they still get paid. This prevents worker frustration in well-mapped zones.
Limitations: The system is entirely dependent on the recency of Street View imagery. If a bike rack was installed yesterday but the Google car drove by two years ago, the system is blind. Future work could integrate multiple imagery sources (Mapillary, OpenStreetCam) or use drone footage for more dynamic auditing.
In conclusion, the Virtual City Explorer marks a shift from "You have to be there" to "You just have to look," providing a powerful blueprint for future smart city infrastructure management.