ICTD 2024

Highway construction work zones play an indispensable role in preserving and enhancing infrastructure for optimal operational efficacy and commuter comfort. Yet, these zones remain critical safety chokepoints within transportation networks, presenting amplified risks of accidents resulting in grievous injuries or fatalities to both construction personnel and drivers. Predominantly, Truck truck-mounted attenuators (TMAs) have been employed to mitigate accident impact severity within these zones. Contemporary assessments discern that most work zone accidents can be attributed to distracted driving and drivers' misjudgment regarding TMA velocities and proximities. Conventional TMA systems necessitate human operators within TMA vehicles to surveil approaching traffic and actuate auditory and visual warnings. Addressing this, our research pioneered an automated TMA alert mechanism hinging on a single monocular vision camera, which not only optimizes accuracy but also presents a cost-efficient solution in terms of deployment and maintenance. Coupled with deep learning algorithms, our architecture continually assesses incoming vehicular behaviors, gauging their velocities and proximities to the TMA. Leveraging a confluence of deep learning methodologies, we discern vehicular intent, forecasting potential collision risk and estimating time-to-collision. Based on these determinations, the system autonomously activates appropriate warning modalities—strobe lights, sirens, or a combination thereof. Simulated evaluations underscore the AutoTMA system's efficacy, illustrating a remarkable 50% reduction in potential accident risk. We envisage our findings catalyzing paradigmatic shifts in work zone safety protocols, striving toward the overarching objective of obliterating work zone-related mishaps.

A Monocular Vision Deep Learning System for Enhanced Highway Work Zone Safety Enforcement

Primary Author(s)

Co-Author(s)

Submitter(s)

Learning Objectives: