Distributed Fiber-Optic Sensing based Single-Lane Abnormal Event Detection in Low-Density Traffic Flow

TL;DR

This paper presents a fiber-optic sensing method to detect single-lane abnormal traffic events by monitoring vehicle lane changes, achieving over 80% accuracy in real-world tests, thus enabling early congestion mitigation.

Contribution

It introduces a novel approach combining vehicle path estimation and spectral analysis of vibrations to identify lane-change maneuvers in low-density traffic.

Findings

81.5% accuracy in vehicle path tracking

83.5% accuracy in lane-change detection

Potential for early abnormal event detection

Abstract

Distributed fiber-optic sensing (DFOS) based traffic flow monitoring systems are a cost-effective wide-area traffic monitoring solution that utilize existing fiber infrastructure along roads. These systems analyse vehicle vibrations and measure average traffic speeds to detect traffic events. However, these systems face difficulties in detecting early signs of non-recurring traffic congestions in low-density traffic flow caused by presence of single-lane abnormal events. This is because average traffic speeds do not decrease quickly in such events. During abnormal events, multiple vehicles perform spontaneous braking and abrupt lane-changes to avoid obstacles on travel lanes. These vehicle behaviours gradually lead to traffic congestion. Thus, frequent lane-change maneuver, performed by multiple vehicles at similar location, may suggest occurrence of congestion-inducing abnormal events. This paper discusses methods to identify and locate occurrence of these single-lane abnormal events by detecting frequent lane-change maneuver along road sections. We first propose a method to locate vehicle positions along a road section and estimate the vehicle path. We then propose a method to detect vehicle lane-change maneuver by monitoring variations in spectral centroid of vehicle vibrations. The evaluation of our proposed methods with real traffic data for two different expressways showed 81.5% accuracy for individual vehicle path tracking and 83.5% accuracy in lane-change event detection. These results suggest that the proposed method has potential for detecting occurrence of single-lane abnormal events in low-density traffic flow so that necessary mitigation measures can be initiated before onset of traffic congestions.