TL;DR
This paper introduces a collaborative sensing platform that leverages mobile and stationary sensors to create high-resolution, real-time environmental monitoring. It then uses this data to offer an "Eco Routing" service, which helps users find routes that minimize exposure to air pollution, demonstrating significant reductions in pollution exposure for cyclists and cars.
What Problem Does This Paper Address?
Urban air pollution is a growing concern, especially in densely populated areas. While some environmental zones and monitoring stations exist, they often lack the granularity and real-time data needed for effective management and public awareness. The authors identify a need for a system that can harness the power of widespread, mobile sensing (e.g., from smartphones) to provide detailed, up-to-date pollution maps and actionable services like optimized routing. This would empower both administrative bodies for better planning and individuals to make healthier choices regarding their travel routes.
Methods
The core of the proposed system is a Collaborative Sensing Platform with a layered architecture:
- Sensor & Storage Layer: Collects data from mobile (e.g., smartphones) and stationary sensors, and stores persistent information like street networks.
- Data Layer: Handles receiving sensor data, adding timestamps, performing basic checks, and partitioning data by location before storage. It also facilitates data exchange between layers.
- Logic Layer: This is where the intelligence resides, comprising three key components:
- Data Interpolation: Uses geostatistical techniques like kriging to estimate pollution levels across areas, even where direct sensor readings are sparse. They employ a quadtree data structure for efficient spatial organization and a quantile split for data distribution.
- Exposure Model: Calculates an "exposure value" for road segments, considering both the pollutant concentration and the time spent on that segment. This is crucial for the Eco Routing service.
- Routing: Integrates the exposure model into routing algorithms.
- Service Layer: Provides user-facing services like Eco Routing and visualization.
The Eco Routing approach specifically uses the calculated exposure values to weight road segments. They propose two metrics: "fastest eco path" (which considers driving time and exposure) and "shortest eco path" (which normalizes exposure by driving time to get an average pollutant concentration per unit length). The A* routing algorithm is used for pathfinding.
For visualization, they propose bitmap representations, 3D models, and direct mapping onto street networks, allowing users to see pollution levels in different formats.
Datasets
The paper does not use real-world sensor data for its primary evaluations. Instead, they generate synthetic sensor input data using a Land-use regression (LUR) model. This adapted model integrates information from OpenStreetMap (street network), traffic density estimates, and the ASTER GDEM dataset (absolute altitude) to simulate NO2 pollution levels. This approach was chosen due to the lack of readily available, large-scale sensor data.
Key Insights and Ideas
- Ubiquitous Sensing for Granular Pollution Data: The fundamental idea that leveraging the vast number of mobile devices can create pollution maps with unprecedented resolution and real-time accuracy is a key takeaway.
- Exposure-Centric Routing: Moving beyond just distance or time, routing based on minimizing actual pollution exposure is a novel and impactful application. This acknowledges that simply traveling faster or shorter might not be healthier if the route is more polluted.
- Kriging for Pollution Estimation: The application of geostatistical interpolation techniques like kriging, combined with a quadtree structure, provides a robust method for estimating pollution levels in areas without direct sensor coverage.
- Cyclists Benefit Most: The results highlight that cyclists, who are often more exposed to ground-level pollution and have less control over their speed, see the most significant benefits from Eco Routing in terms of reduced pollution exposure.
Key Results
- Interpolation Performance: The comparison between raster and point interpolation methods showed they perform similarly, with the mean ratio (q) close to 1.0 and a low standard deviation, indicating good agreement. The run time for point interpolation decreases significantly with larger distances between measuring points, suggesting a sweet spot around 45m. Raster interpolation run time scales better with larger raster sizes.
- Eco Routing for Cars (Shortest Path): Traditional shortest path routing outperformed Eco Routing, leading to an increase in average driving time (32.2%) and pollution exposure (6.7%), despite efforts to reduce it. This suggests that prioritizing distance over pollution can be detrimental to health.
- Eco Routing for Cars (Fastest Path): The fastest eco path for cars resulted in a significant reduction in pollution exposure (8.6%) with a moderate increase in driving distance (1.9%) and driving time (10.3%).
- Eco Routing for Bicycles: This showed the most promising results, with a 12.6% lower average pollution exposure compared to traditional fastest path routing. This was achieved with only a moderate increase in driving distance and time (7.1%). Notably, 10% of simulated routes saw a 25% reduction in exposure, and 30% saw at least a 17% reduction.
Limitations and Future Work
- Synthetic Data: The evaluations were based on simulated pollution data generated by an LUR model, not real-time sensor readings. This limits the direct applicability of the results to real-world scenarios.
- Static Edge Weights: The current Eco Routing approach uses static edge weights calculated beforehand. For dynamic traffic and pollution changes, routes would need to be recalculated after defined timeouts.
- Future Work:
- Prototypical Sensor Board: Developing and testing a low-power sensor board connected to a smartphone for real-world data collection.
- Additional Services: Implementing services like dynamic pricing systems (e.g., for city tolls or bike rentals) based on current environmental conditions.
- Data Collection Cost: Addressing the cost-intensiveness of continuous data collection (energy, traffic, money).
- Cloud Computing: Enhancing system performance through cloud technologies.
Summary updated