Automatic Traffic Flow Fundamental Diagram Fitting Method based on Trajectory Data

Fundamental diagram (FD) fitting using loop detector data is limited by its coarse granularity and inability to capture spatiotemporal dynamics. Although emerging trajectory datasets can overcome these limitations by offering precise vehicle movement information, their potential for FD fitting has not been fully explored. This paper proposes a dynamic partitioning method for the time-space diagram (TSD) that extracts quasi-steady states to enable reliable FD estimation. A state homogeneity metric, unit weighted signed area (UWSA), is defined and integrated into a multi-regimes stochastic FD fitting model, where regime boundaries are identified via kernel density estimation. Compared with traditional methods based on loop detectors, the proposed method significantly increases the volume of usable data and enables continuous FD estimation in the TSD, thereby facilitating global assessment of spatial heterogeneity. Based on Zen Traffic Data, FD fitting and spatial analysis of capacity demonstrate that: (a) the capacity of passing lanes is on average 39.03% higher than that of driving lanes, indicating obvious lane-level discrepancies; (b) both lanes exhibit noticeable capacity drops, with averages of 10.90% for passing lanes and 21.71% for driving lanes; (c) segment-level capacity varies significantly with roadway conditions, and locations such as curves or junctions exhibit the most pronounced capacity degradation. These results confirm the performance of the proposed method for FD fitting and the ability to continuously analyze traffic flow.