Core Concepts
A real-time framework is proposed to evaluate pedestrian's potential risk at non-signalized intersections using a novel surrogate safety measure called Predicted Post-Encroachment Time (P-PET), which is derived from deep learning models that predict the arrival time of pedestrians and vehicles.
Abstract
The key highlights and insights of this content are:
The authors propose a framework that combines computer vision technologies and predictive models to evaluate pedestrian's potential risk in real-time at non-signalized intersections.
A novel surrogate safety measure called Predicted Post-Encroachment Time (P-PET) is introduced, which is derived from deep learning models that can predict the arrival time of pedestrians and vehicles. P-PET is used to accurately and interpretably evaluate the potential risk of pedestrians.
Pedestrians are classified into three categories - kids, adults, and cyclists - and specific evaluation rules are applied to each category to enhance the effectiveness and reliability of the risk evaluation.
The framework is tested at a real-world non-signalized intersection in Sejong City, South Korea. The results demonstrate the framework's ability to effectively identify potential risks using P-PET and its improved performance in risk evaluation across different pedestrian categories.
The framework addresses key research gaps in existing prediction-based potential risk evaluation studies, including the lack of a real-time application framework, the need for accurate and explainable safety indicators, and the lack of tailored evaluation criteria for different pedestrian categories.
The video processing pipeline utilizes computer vision techniques like object detection, tracking, segmentation, and perspective transformation to extract trajectory data for vehicles, adults, kids, and cyclists.
Deep learning models like LSTM, GRU, and Transformer are employed to predict the arrival time of pedestrians and vehicles, which is crucial for calculating P-PET.
The potential risk evaluation algorithm leverages P-PET and applies specialized criteria for different pedestrian categories to assess their potential risk at the intersection.
Stats
The following sentences contain key metrics or important figures used to support the author's key logics:
"Pedestrians are classified into kids, adults, and cyclists to tailor specific evaluation rules to each category of pedestrians, thus enhancing the effectiveness and reliability of risk evaluation."
"The results at a real-field, non-signalized intersection show the effectiveness of the proposed framework and illustrate the computational time costs across different components, demonstrating its potential in real-time application."
Quotes
"A framework with computer vision technologies and predictive models is proposed to evaluate pedestrian's potential risk in real time."
"A novel predicted surrogate safety measure, Predicted Post-Encroachment Time (P-PET), is introduced to accurately and interpretably evaluate potential pedestrian risk."