Crowdsourcing the Green Light: Precise SPaT Estimation via Queue Physics

TL;DR

Researchers have developed a system that uses low-frequency GPS data from city buses to predict traffic signal timings. By incorporating a mathematical model of "queue dissipation"—how long it takes for a line of cars to start moving—they can accurately estimate when a light turns green, even during heavy San Francisco rush hour traffic.

Background: The I2V Connectivity Gap

Vehicle-to-Infrastructure (V2I) communication is the backbone of fuel-efficient cruise control and collision avoidance. However, most cities don't share their Traffic Management Center (TMC) data publicly. While crowdsourcing is a popular alternative, previous attempts struggled with a major paradox: the data is most needed during traffic jams, but traffic jams (queues) "pollute" the data because vehicles don't move the moment the light turns green.

The Problem: The "Queue Delay" Distortion

In earlier studies, researchers simply ignored data from vehicles stuck in traffic because it was too hard to tell if a bus stopped because of a red light or because of the five cars in front of it. This lead to:

1. Low Data Utilization: Up to 90% of useful samples were discarded.
2. Rush Hour Inaccuracy: Signal clocks drift over the day; excluding rush hour data means the system loses track of the "true" timing precisely when it's most needed.

Methodology: Modeling the "Human" Start-up Wave

The core innovation lies in how the authors handle the delay between the start of green ($t_{SoG}$) and the moment a vehicle actually starts moving ($t_{start}$).

1. Reconstructing Trajectories from Sparse Data

Since buses only report location every 90 seconds (low frequency), the system "interpolates" the movement. It fits sparse GPS points into a velocity-vs-time model that accounts for deceleration, idling in a queue, and acceleration.

2. The Queue Clearance Model

Instead of assuming a constant delay, the authors use a Multiple Linear Regression on historical data to define an exponential headway model: $\Delta t_{clearance}=hN+{\Delta }_1{\sum }_{n=1}^N{e}^{-(n-1)}$ Where:

• $h$ is the saturation headway (time between cars).
• $N$ is the position in the queue.
• ${\Delta }_1$ is the extra delay for the first driver's reaction.

By knowing the bus's distance from the stop-bar (position $N$), the algorithm can "subtract" the expected queue delay to find the exact moment the light changed.

Experiments: Real-World San Francisco Validation

The system was tested using public bus feeds in San Francisco. Unlike simulation-heavy papers, this work confronted real-world "noise" like bus stops and varying street widths.

• Green-Initiation Accuracy: During heavy traffic, the "Queue-Aware" model (solid red line in Fig 16) significantly outperformed the "Queue-Ignorant" model (dashed line).
• Robustness: Even with buses arriving only every 5-10 minutes, the moving average filter effectively tracked signal clock drifts and schedule changes (e.g., switching from off-peak to rush-hour timings).

Critical Insight & Conclusion

This paper demonstrates a significant shift from "big data" to "smart physics." Instead of requiring high-frequency 1Hz data from every car on the road, the authors show that with a deep understanding of traffic flow theory, we can extract high-fidelity infrastructure data from low-fidelity public sources.

Limitations

• Pre-timed focus: Currently limited to signals with fixed schedules. Actuated signals (that change based on real-time demand) remain a much harder "moving target."
• Bus Stop Noise: Urban buses stop for passengers, which can mimic a red-light stop. The current filter relies on "cyclic periodicity" to tell the difference, but high-density bus stops still introduce outliers.

Final Takeaway

For developers of Connected Vehicle apps, this provides a blueprint for a "software-only" infrastructure rollout. By deploying these back-end algorithms, a city can enable SPaT features for all citizens using nothing but their existing public transit GPS feed.