Scholarlink

<p>Safety in transportation engineering, especially at unsignalized intersections, is a critical concern traditionally assessed through accident records. However, the existing method faced challenges in emerging nations like India, where accidents are prevalent. The study presents a novel approach utilizing the Gumbel and Scaled Unit-based Deep Radial Basis Function Network (GSU-DRBFN) for evaluating safety at unsignalized intersections. By assessing the safety measures in the unsignalized intersections through various processes of the proposed work, the features of the unsignalized intersections can be efficiently learned and the possibility of accidents can be accurately predicted. Such prior predictions of accidents in the unsignalized intersections contribute to the transportation engineering field in planning and designing road transportation accordingly so that the public can relish a safer travelling experience. For evaluating the safety at unsignalized intersections, initially, videos are transformed into individual image frames, and motion estimation is applied for reducing temporal redundancy. The frames undergo enhancement using the Range-based Contrast Limited Adaptive Histogram Equalization (RCLAHE) algorithm. The enhanced frames are then subjected to foreground and background extraction, separating the foreground and background frames, and features are extracted. Moving and non-moving objects are clustered using the Ratio Fuzzy C-Means (RFCM) technique, and surrogate measures are extracted. A threshold is determined by utilizing the peak hour threshold method in extreme value theory for the Post Encroachment Time (PET) that, in conjunction with background features, is utilized in the prediction phase of the GSU-DRBFN technique for analyzing conflict severity. Experimental trials are conducted to validate the methodology, demonstrating the effectiveness of the proposed model in comparison to other algorithms, namely Deep Radial Basis Function Network (DRBFN), Deep Neural Network (DNN), Convolutional Neural Network (CNN), and Artificial Neural Network (ANN), that use Deep Learning (DL) and artificial intelligence. The proposed GSU-DRBFN approach achieves a maximum accuracy of 97.85%, outperforming the prevailing classifier by approximately 2.35%. For all the quantitative metrics, such as precision and specificity, the proposed approach demonstrates accurate performance in traffic conflict scenario prediction. Thus, the performance gain exhibits the significance of the proposed approach in improving proactive safety assessment at unsignalized intersections, enabling improved transportation planning.</p> Graphical abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Proactive safety assessment of unsignalized intersections using GSU-DRBFN-based predictive modeling approach

  • 期刊论文

  • 发表日期：

    2026年7月29日

查看全文
  • Dungar Singh,
  • Pritikana Das,
  • Indrajit Ghosh

摘要

Safety in transportation engineering, especially at unsignalized intersections, is a critical concern traditionally assessed through accident records. However, the existing method faced challenges in emerging nations like India, where accidents are prevalent. The study presents a novel approach utilizing the Gumbel and Scaled Unit-based Deep Radial Basis Function Network (GSU-DRBFN) for evaluating safety at unsignalized intersections. By assessing the safety measures in the unsignalized intersections through various processes of the proposed work, the features of the unsignalized intersections can be efficiently learned and the possibility of accidents can be accurately predicted. Such prior predictions of accidents in the unsignalized intersections contribute to the transportation engineering field in planning and designing road transportation accordingly so that the public can relish a safer travelling experience. For evaluating the safety at unsignalized intersections, initially, videos are transformed into individual image frames, and motion estimation is applied for reducing temporal redundancy. The frames undergo enhancement using the Range-based Contrast Limited Adaptive Histogram Equalization (RCLAHE) algorithm. The enhanced frames are then subjected to foreground and background extraction, separating the foreground and background frames, and features are extracted. Moving and non-moving objects are clustered using the Ratio Fuzzy C-Means (RFCM) technique, and surrogate measures are extracted. A threshold is determined by utilizing the peak hour threshold method in extreme value theory for the Post Encroachment Time (PET) that, in conjunction with background features, is utilized in the prediction phase of the GSU-DRBFN technique for analyzing conflict severity. Experimental trials are conducted to validate the methodology, demonstrating the effectiveness of the proposed model in comparison to other algorithms, namely Deep Radial Basis Function Network (DRBFN), Deep Neural Network (DNN), Convolutional Neural Network (CNN), and Artificial Neural Network (ANN), that use Deep Learning (DL) and artificial intelligence. The proposed GSU-DRBFN approach achieves a maximum accuracy of 97.85%, outperforming the prevailing classifier by approximately 2.35%. For all the quantitative metrics, such as precision and specificity, the proposed approach demonstrates accurate performance in traffic conflict scenario prediction. Thus, the performance gain exhibits the significance of the proposed approach in improving proactive safety assessment at unsignalized intersections, enabling improved transportation planning.

Graphical abstract