Development of Reliable Models of Signal-Controlled Intersections

Glushkov, Alexandr; Shepelev, Vladimir · 2021 · Crossref

DOI: 10.2478/ttj-2021-0032

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of developing reliable mathematical models for signal-controlled intersections, which often exhibit significant variability in traffic capacity due to heterogeneous conditions and random factors. The authors argue that treating all intersections as a single group leads to low-quality predictive models. To improve management efficiency, they propose dividing intersections into homogeneous groups using clustering methods, allowing for the creation of generic, cluster-specific management algorithms. The study utilizes data collected from 25 intersections in Chelyabinsk, Russia, via street video surveillance cameras processed by the YOLOv3 convolutional neural network. The dataset includes 20 recorded parameters, encompassing fixed factors (e.g., intersection geometry, road surface quality) and variable factors (e.g., weather, pedestrian flow, traffic intensity). The researchers first constructed a linear regression model for the entire set of intersections, which yielded an R-squared value of 0.672, below the acceptable threshold of 70%. Consequently, they applied statistical clustering to divide the intersections into two homogeneous groups: one with 14 intersections and another with 11. The clustering was driven primarily by differences in pedestrian flow characteristics and flow discontinuity. Linear regression models were then developed for each cluster. Additionally, a fuzzy logic model using Gaussian membership functions was constructed for the first cluster to visually assess the influence of pedestrian flow intensity and continuity on traffic capacity. The results demonstrate that clustering significantly improves model reliability. The linear regression models for the two clusters achieved R-squared values of 81.7% and 74.0%, respectively, both exceeding the 70% quality threshold and outperforming the aggregate model. While the statistical confidence of the cluster models was slightly lower than that of the complete model, the authors attribute this to the small sample size, noting that larger datasets would likely enhance confidence further. The fuzzy logic analysis revealed that pedestrian flow density has the most substantial impact on intersection traffic capacity, while flow discontinuity acts as a random disturbing factor that does not radically alter the general trend. The study concludes that dividing signal-controlled intersections into homogeneous groups is a necessary step for developing effective, autonomous management decisions for urban transport networks. By identifying statistically significant differences between clusters, traffic management systems can apply tailored control strategies rather than generic ones. The authors suggest that expanding data collection through neural networks will further improve the quality of modeling and prediction, supporting more adaptive and efficient traffic light control systems.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• gap acceptance