Advanced operations focused on connected vehicles/infrastructure (CVI-UTC).
archive: archived pipeline: cataloged verified
Get this paper ↗ (full text — opens at the source; we link to it, we don't host it)
Summary
This paper addresses the challenge of identifying infrastructure safety "hot spots" more proactively and accurately than traditional methods, which rely on police-reported crash data. Traditional approaches are reactive, requiring significant time for statistically significant crash accumulations, and often suffer from location inaccuracies. The authors propose using Connected Vehicle (CV) data, specifically Basic Safety Messages (BSM), to detect not only crashes but also "near-crashes"—events involving rapid evasive maneuvers. By analyzing kinematic data from these events, transportation agencies could identify safety issues faster and with greater precision. The study aims to develop and test algorithms capable of detecting crashes and near-crashes using vehicle kinematic data without relying on Time-to-Collision (TTC) metrics, which are currently infeasible due to low V2V market penetration. The research utilized data from the Virginia Tech Transportation Institute’s 100-Car Naturalistic Driving Study (NDS). The training dataset consisted of 13 usable crash events, manually labeled using video verification. Two additional datasets were used for testing: a "normal driving" set of 14 trips with no crashes or near-crashes to measure false positive rates, and a larger dataset containing 68 crashes and 760 near-crashes to assess sensitivity. The authors tested three modeling techniques against simple acceleration thresholds: Multivariate Adaptive Regression Splines (MARS), Classification and Regression Trees (CART), and a novel pattern matching algorithm. The pattern matching approach involved defining five baseline acceleration profiles for common driving actions (e.g., accelerating from a stop, braking) and using a sliding window to calculate Euclidean distances between observed data and these baselines. Events that did not match any baseline within a defined threshold were flagged as potential crashes or near-crashes. The results indicated that while conclusive results were not achieved due to data limitations, the models showed potential. The pattern matching approach outperformed simple acceleration thresholds. In a crash-only test set, it identified nearly 70% of crashes, and in a near-crash-only test set, it identified 74% of events. On the training set, the pattern matching method identified more crashes than threshold methods without increasing the number of false positives. The CART and MARS models also demonstrated the ability to identify the majority of the 13 known crashes, though they produced varying amounts of false positives. The study highlighted that simple thresholds often lead to trade-offs between missing lower-severity crashes and generating excessive false alarms, whereas pattern recognition could better distinguish normal driving variations from safety-critical events. The significance of this work lies in its demonstration that CV kinematic data can serve as a viable surrogate for traditional crash data in identifying infrastructure safety issues. The findings suggest that near-crash data, which are more frequent than actual crashes, can improve the precision and speed of hot spot identification. Although the study did not fully validate the methodology on a live field test bed due to data availability constraints, it established a foundational approach for future implementation. The CVI-UTC concluded that it is prepared to apply this methodology to data collected on its Northern Virginia Connected Vehicle test bed, potentially enabling a more proactive Roadway Safety Management Process.
Key finding
The pattern matching algorithm identified nearly 70% of crashes and 74% of near-crashes in test sets, outperforming simple acceleration thresholds.
Methodology
dataset
Sample size: 13
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via bulk_ingest_rosap on 2026-05-23 (6 acquisition events logged).
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | rosap | — | — | 2 | 2026-05-23 |
| archive | success | — | — | — | 1 | 2026-05-23 |
| extract | success | cached | — | — | 2 | 2026-06-10 |
| clean | success | — | — | — | 1 | 2026-06-01 |
| chunk | success | — | — | — | 1 | 2026-06-01 |
| embed | success | — | — | — | 1 | 2026-06-02 |
| enrich | success | — | — | — | 1 | 2026-05-23 |
| promote | success | — | — | — | 1 | 2026-05-23 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 3 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 19 | 2026-06-11 |
| verify | partial | — | — | — | 2 | 2026-06-10 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified_with_issues.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
- naturalistic crash near crash
- telematics crash prediction
- incidence prevalence
- induced exposure
- crash typology
- causation analyses
Information type
What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).
- Empirical Findings: crash risk outcomes
- Methodological Resource: dataset resource, validation psychometrics