Recognition of Intersection Traffic Regulations from Crowdsourced Data

Zourlidou, Stefania; Sester, Monika; Hu, Shaohan · 2022 · OpenAlex-citations

DOI: 10.3390/ijgi12010004

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 automatically identifying traffic regulations at intersections, such as traffic lights and stop signs, using crowdsourced GPS trajectory data. The authors motivate this work by noting that while traffic controls significantly impact travel time, fuel consumption, and emissions, this information is largely absent from digital maps. Traditional mapping methods are costly and slow to update, creating a need for scalable, inexpensive solutions. The study aims to determine how collective vehicle behavior can be leveraged to classify intersection control types efficiently. The proposed method analyzes GPS traces to detect stopping and deceleration episodes using a modified clustering algorithm. These episodes categorize vehicle crossings into four behavioral classes: free flow, deceleration without stopping, single stopping event, and multiple stopping events. The study evaluates three classification models: a dynamic model using features extracted from trajectories (speed, stop, and deceleration statistics), a static model using features from OpenStreetMap (e.g., road hierarchy, distance), and a hybrid model combining both. For each model, the authors test two variants: a "one-arm" model using only features from the specific intersection arm being classified, and an "all-arm" model incorporating features from neighboring arms of the same intersection. The methodology was validated on three distinct datasets to assess performance across different cities and trajectory densities. The results demonstrate that the hybrid model, which integrates static map data with dynamic trajectory features, outperforms both the purely dynamic and purely static models. Furthermore, the "all-arm" models consistently achieved higher accuracy than the "one-arm" models, indicating that contextual information from neighboring intersection arms is valuable for classification. The all-arm hybrid models achieved classification accuracies between 95% and 97%. The study also highlights that existing literature often neglects the use of neighboring arm features in dynamic models and lacks standardized benchmarks, making direct comparison difficult. The significance of this work lies in providing a robust, cost-effective framework for updating traffic regulation maps using ubiquitous GPS data. By proving that hybrid approaches and contextual arm features improve accuracy, the paper offers a scalable solution for Smart City applications, such as accurate travel time estimation and emission modeling. The findings suggest that leveraging collective driver behavior, combined with existing map data, can effectively fill gaps in geographic information systems without the need for expensive surveying equipment or specialized hardware.

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.

• crash typology