A Study on a Deep Learning Model-based Method for Complex Event Intersection Risk Prediction and Hotspot Visualization Considering Trajectory Uncertainty 


Vol. 14,  No. 10, pp. 838-849, Oct.  2025
https://doi.org/10.3745/TKIPS.2025.14.10.838


PDF
  Abstract

This study proposes a composite event-based collision risk prediction framework for complex, high-uncertainty urban intersections involving multiple interacting road users. Unlike conventional approaches based on single-point estimates or isolated risk signals, the proposed method integrates major unsafe behaviors (sudden braking, lane changing, close following), quantitative risk indices (TTC, PET), and overlap of predicted uncertainty intervals (IoU) to assess risk. The analysis leverages multi-vehicle trajectories extracted from three real intersections (Sites A, B, C) in the DRIFT dataset, with risk scores normalized and visualized spatially, and each location classified into four risk levels. Empirical results show that high-risk hotspots are strongly concentrated at specific locations—merging zones, crossings, and bottlenecks—where composite events frequently and repeatedly occur, resulting in rapid risk accumulation. The composite signal criteria substantially improve precision and F1-score over traditional single-feature approaches, validating practical utility for real-time risk monitoring and targeted intervention. However, a limitation is that event score weighting currently relies on heuristics; future studies should calibrate using accident statistics and domain expertise. This framework provides actionable support for scientific intersection.

  Statistics
  Cite this article

[IEEE Style]

M. J. Hong and B. S. Kim, "A Study on a Deep Learning Model-based Method for Complex Event Intersection Risk Prediction and Hotspot Visualization Considering Trajectory Uncertainty," The Transactions of the Korea Information Processing Society, vol. 14, no. 10, pp. 838-849, 2025. DOI: https://doi.org/10.3745/TKIPS.2025.14.10.838.

[ACM Style]

Min Ju Hong and Byeong Soo Kim. 2025. A Study on a Deep Learning Model-based Method for Complex Event Intersection Risk Prediction and Hotspot Visualization Considering Trajectory Uncertainty. The Transactions of the Korea Information Processing Society, 14, 10, (2025), 838-849. DOI: https://doi.org/10.3745/TKIPS.2025.14.10.838.

Forms

Search
(IN TITLE, AUTHOR, ABSTRACT,KEYWORDS)

Advanced Search ▼

POPULAR KEYWORDS
(TOP 10 KEYWORDS)

Recent Publications
(LAST 3 YEARS)

Old Journals

Indexing

All publications of TKIPS is indexed in DOI, EBSCO, Google Scholar, Crossref, and CrossCheck.

TKIPS is also selected as the Journal for Accreditation by NRF (National Research Foundation of Korea).

Related Journals