PERFORMANCE OPTIMIZATION OF ADAPTIVE TRAFFIC LIGHTS USING MACHINE VISION

Shepelev, Vladimir; Almetova, Zlata; Moor, Aleksandr; Bersteneva, Valeria · 2020 · Crossref

DOI: 10.14529/em200119

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 optimizing traffic throughput at signalized intersections by integrating machine vision and neural networks into adaptive traffic light control systems. Motivated by the increasing urbanization and the limitations of existing static or simplified adaptive traffic models, the authors propose a new Intelligent Infrastructure-to-Pedestrian (I2P) approach. The primary goal is to maximize vehicle capacity, particularly for right-turning traffic, while minimizing negative impacts on pedestrian safety and flow. The study highlights that current software solutions often fail to account for real-time dynamic constraints, leading to inefficiencies in urban road networks. The research was conducted at a heavily congested intersection in Chelyabinsk, Russia, characterized by significant traffic jams between 7:30 AM and 7:30 PM. The methodology involved the dynamic monitoring of both vehicular and pedestrian traffic using street surveillance cameras equipped with neural networks. These networks were trained to interpret video streams in real-time, counting vehicles (categorized into six types) and pedestrians for each lane. The authors analyzed factors reducing infrastructure efficiency, such as conflicts between right-turning vehicles and pedestrians, and the presence of public transport. Theoretical calculations of lane capacity were compared with actual data collected via cameras and validated through microscopic simulation using the PTV Vissim software. The findings reveal that the lane allowing straight and right-turn movements (1RS_N) has the lowest throughput due to pedestrian conflicts and the need to yield. The proposed adaptive algorithm dynamically adjusts pedestrian signal timings based on real-time traffic density. By reducing the pedestrian green phase duration only when necessary to accommodate high vehicular demand, the system optimizes flow. Specifically, every two seconds reduced from the pedestrian phase increases lane capacity by 28–30 equivalent vehicles per hour. Simulation results indicated that this approach increases the throughput of the right-turn/straight lane by 8–10% under minimal implementation and potentially up to 50% under optimal conditions. The simulated throughput was found to be 4% higher than theoretical calculations, validating the model's accuracy. The significance of this work lies in its demonstration that machine vision can effectively bridge the gap between theoretical traffic models and real-world dynamics. By leveraging real-time data to adjust signal timings, the proposed "smart traffic light" system offers a scalable solution for urban intersections without requiring costly physical infrastructure expansions. The study concludes that integrating neural network-based video analytics into traffic control systems can significantly enhance intersection efficiency and reduce congestion, providing a practical framework for smart city development.

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.

• vru facing ehmi