Empirical calibration inputs for micro-simulation of heterogeneous, non-lane-based traffic

Abstract

Urban arterials in developing countries carry highly heterogeneous, non-lane-based traffic that fundamentally differs from the lane-disciplined streams for which conventional micro-simulation tools were designed. The calibration of lateral movement parameters in simulation platforms requires locally validated empirical data on where each vehicle class positions itself across the carriageway and what lateral clearances are accepted in practice. This study analyses trajectory data from a four-lane undivided urban arterial in Sri Lanka using a YOLOv11 and ByteTrack computer vision pipeline with homography-based spatial calibration, achieving positional accuracy below 0.15 m root mean square error. Two metrics are proposed: lateral position distributions (Metric A) across 11 vehicle classes and pairwise edge-to-edge lateral clearance distributions (Metric B) for major vehicle-class combinations. Results reveal pronounced area-based carriageway utilization with no effective lane discipline. Motorcycles and three-wheelers concentrate in kerb-adjacent zones, while motorcycles exhibit a median encroachment frequency of 17.67%. Motorcycle-involved vehicle pairs consistently record the smallest lateral clearances across all speed categories, with motorcycle–motorcycle and motorcycle–two-wheeler encounters averaging below 0.95 m. These findings provide a validated local parameter set for micro-simulation calibration and carry direct implications for lane-allocation policy and road safety interventions targeting two-wheeler interactions on heterogeneous traffic arterials.