Video-Based Object Recognition for Identifying Vehicle Distances Needed in Bridge Load Evaluations

Ralbovsky, Marian; Aleksa, Michael; Hula, Andreas · 2025 · Crossref

DOI: 10.1007/978-3-032-04774-8_42

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the critical need for accurate data on vehicle distance distributions in congested traffic, a key parameter for evaluating the reliability and maximum traffic load effects on bridges. Current bridge design standards, such as the EuroCode, rely on traffic measurements from the 1980s, while modern simulations often use simplified assumptions like constant or uniform distance distributions. The authors aim to expand the limited existing datasets by employing an innovative video-based measurement method to capture realistic vehicle spacing behaviors, particularly in congestion scenarios where maximum load effects typically occur. The researchers utilized the “Mobility Observation Box” (MOB), a compact, self-sustaining camera system equipped with machine learning algorithms for automatic vehicle detection and trajectory tracking. Video data was collected at two distinct highway sites with different traffic characteristics: Site 1 featured predominantly free-flowing traffic, while Site 2 exhibited significant congestion and a higher proportion of trucks. The analysis involved processing video frames to identify bounding boxes for vehicles, which were then mapped to meter values on the driving lane to calculate inter-vehicle distances. Two evaluation methods were applied depending on camera angles: one calculated distances directly from bounding box positions, while the other estimated vehicle lengths and time gaps based on crossing reference lines. To separate congestion behavior from flowing traffic, the data was analyzed across several velocity ranges. The results demonstrated that vehicle spacing behavior varies significantly by speed. In flowing traffic conditions (30–80 km/h) at Site 1, vehicles maintained consistent time gaps rather than fixed distances, leading the authors to model spacings using time gaps. In contrast, for congestion scenarios (velocities below 20 km/h), net vehicle distances were analyzed. The study found that vehicle distances in congestion followed a roughly lognormal distribution, contrary to some previous literature suggestions. Furthermore, fitting the data with a bimodal distribution (combining Weibull and lognormal distributions) provided a better match to the observed data than single lognormal distributions, although the difference was not drastic. Site 2 provided more robust data for congestion analysis due to its higher frequency of slow-moving traffic. The significance of this work lies in validating video-based object recognition as a feasible and accurate method for gathering detailed traffic data for bridge load assessments. By demonstrating that bimodal distributions better capture vehicle spacing in congestion, the study offers improved input parameters for traffic flow simulations used in structural engineering. The authors conclude that while the current dataset is limited to two sites, the methodology is scalable and can be applied to a wider selection of road sites to derive general recommendations for modeling vehicle distances, ultimately enhancing the precision of bridge reliability evaluations.

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.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• exposure measurement