Congestion Identification at Urban Intersections using FCD

Traditional congestion detection relies on sparse fixed detectors. FCD offers wide coverage but high noise, requiring map-matching and state estimation before use.

Identify congestion states and propagation directions at urban intersections from massive FCD, supporting signal optimization.

Map-matched GPS with an HMM, aggregated intersection speed in 5-min bins; built 2D speed-density features, compared K-Means vs DBSCAN, then inferred propagation chains via lagged correlation.

A data-driven view revealed the "contagiousness" of congestion—failure at one intersection propagates downstream at ~12 km/h. This inspired me to couple simulation with data for closed-loop validation.