Scholarlink

<p>Traffic streams in developing economies like India often exhibit disorganised behaviour due to vehicle heterogeneity, lack of lane discipline, and complex interactions, posing challenges for accurate modelling. This study aims to address this gap by performing regression-based macroscopic and microscopic analysis of high-resolution Unmanned Aerial Vehicles (UAVs) based data from a disorderly stream, with the primary contribution being the development of tailored traffic flow models that capture randomness and variability in heterogeneous conditions. The analysis focuses on high-resolution vehicular trajectory data from an eight-lane highway on Delhi’s outskirts, collected via UAVs and processed using the fully automated DataFromSky (DFS) software, yielding 8,938,920 data points at 10 frames per second (FPS). The research enhances modelling accuracy for disorderly streams by extracting macroscopic (speed, density, flow) and microscopic (longitudinal spacing, time gaps) parameters, fitting deterministic and machine learning (ML) models, and developing stochastic fundamental diagrams (SFDs). Popular deterministic models were fitted using Least Squares (LSM) and Weighted Least Squares (WLSM). Unweighted models showed higher R² and lower global RMSE due to uncongested-state data-points dominance in the sample. The weighted approach redistributed influence across the density range, giving a more balanced speed–density representation. Among the fitted models, the Underwood model performed most consistently. To capture disorderly traffic’s randomness, Artificial Neural Networks (ANN) and Random Forest (RF) models were also applied, attaining R² comparable to Greenshields but below Underwood, reflecting broader variability. The study further develops SFDs to assess speed variability, support microscopic analysis, and enable car-following model calibration (e.g., Modified GHR, optimal velocity), facilitating 85th percentile speed estimation, traffic control, safety, and resilience, while trajectories enhance microscopic applications, all amplified by ML’s scalability with the large dataset. Limited to longitudinal analysis and single-regime models, the study suggests future exploration of multi-regime models, ML optimisation, and lateral behaviour to fully leverage the data.</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

Regression-based Macroscopic and Microscopic Analysis of UAV-Based Data of a Disorderly Stream

  • 期刊论文

  • 发表日期：

    2026年4月27日

查看全文
  • Shekhar Singh,
  • Surya H. Ravikumar,
  • Akhilesh Kumar Maurya

摘要

Traffic streams in developing economies like India often exhibit disorganised behaviour due to vehicle heterogeneity, lack of lane discipline, and complex interactions, posing challenges for accurate modelling. This study aims to address this gap by performing regression-based macroscopic and microscopic analysis of high-resolution Unmanned Aerial Vehicles (UAVs) based data from a disorderly stream, with the primary contribution being the development of tailored traffic flow models that capture randomness and variability in heterogeneous conditions. The analysis focuses on high-resolution vehicular trajectory data from an eight-lane highway on Delhi’s outskirts, collected via UAVs and processed using the fully automated DataFromSky (DFS) software, yielding 8,938,920 data points at 10 frames per second (FPS). The research enhances modelling accuracy for disorderly streams by extracting macroscopic (speed, density, flow) and microscopic (longitudinal spacing, time gaps) parameters, fitting deterministic and machine learning (ML) models, and developing stochastic fundamental diagrams (SFDs). Popular deterministic models were fitted using Least Squares (LSM) and Weighted Least Squares (WLSM). Unweighted models showed higher R² and lower global RMSE due to uncongested-state data-points dominance in the sample. The weighted approach redistributed influence across the density range, giving a more balanced speed–density representation. Among the fitted models, the Underwood model performed most consistently. To capture disorderly traffic’s randomness, Artificial Neural Networks (ANN) and Random Forest (RF) models were also applied, attaining R² comparable to Greenshields but below Underwood, reflecting broader variability. The study further develops SFDs to assess speed variability, support microscopic analysis, and enable car-following model calibration (e.g., Modified GHR, optimal velocity), facilitating 85th percentile speed estimation, traffic control, safety, and resilience, while trajectories enhance microscopic applications, all amplified by ML’s scalability with the large dataset. Limited to longitudinal analysis and single-regime models, the study suggests future exploration of multi-regime models, ML optimisation, and lateral behaviour to fully leverage the data.