This post is also available in:
עברית (Hebrew)
Traditional efforts to improve Pedestrian safety often rely on police-reported crash data, which can be limited and problematic. To address this gap, researchers at Queensland University of Technology (QUT), Australia, have developed an innovative machine learning model that predicts pedestrian crashes and their severity levels without using crash data. Their groundbreaking study, published in Communications in Transportation Research, could transform the way pedestrian safety measures are implemented.
According to TechXplore, the QUT team collected extensive video data of pedestrian movements at signalized intersections in Brisbane, Queensland, to analyze pedestrian behavior and interactions with vehicles. The goal was to estimate pedestrian crash frequency and severity using computer vision and machine learning techniques, without relying on actual crash records, which are often sparse and influenced by reporting biases.
Fizza Hussain, a researcher at QUT’s School of Civil and Environmental Engineering, explained that the hybrid model estimates pedestrian crash frequency by severity levels. Using machine learning, extreme vehicle-pedestrian interactions are identified and modeled through extreme value theory, which helps distinguish severe from non-severe crashes.
The study’s results were impressive, with the model accurately predicting the frequency of severe and non-severe pedestrian crashes. Over a five-year observation period, the model’s predictions closely matched the actual data, offering a new method to predict crash risks using just a week’s worth of traffic movement data. The study provides evidence that crash risks can be predicted with remarkable accuracy, without waiting for years of crash data. It’s a game-changer for developing tailored countermeasures.
Machine learning plays a pivotal role in this approach. By analyzing pedestrian-vehicle interactions, the model’s crash risk predictions were three times more accurate than conventional methods. This breakthrough allows road authorities to proactively identify high-risk areas and implement targeted safety measures based on predicted severity levels.
This research sets the stage for a new era in pedestrian safety, where machine learning models can offer timely insights and help prevent pedestrian accidents before they happen.
