A GT 3D Bounding-Box-Based Framework for Geometric Quantification and Spatial Validation of Vehicle Frontal Blind Spots
Year: 2026
Type: article
Abstract: Vehicle-induced frontal blind spots can restrict a driver’s direct visibility of pedestrians and other vulnerable road users, and their spatial extent depends on vehicle geometry and driver-viewpoint conditions. Conventional approaches based on category-level representative dimensions may not preserve geometric variation among individual vehicles, whereas detailed visibility models typically require vehicle-specific computer-aided design models and driver accommodation information. This study proposes a geometric framework that uses ground-truth (GT) three-dimensional (3D) bounding boxes from the TUMTraf roadside perception dataset as an intermediate representation for frontal blind-spot quantification and spatial placement. Vehicle-specific GT width is used to estimate the blind-spot ground area and volume, while bounding-box position, dimensions, and yaw are used to spatially anchor the resulting geometry. A mean-width baseline comparison, sensitivity analyses of representative visibility parameters, and calibration-based camera projection were conducted to evaluate complementary aspects of the framework. Replacing vehicle-specific GT widths with the mean-width baseline resulted in a mean absolute percentage error of 5.8% and a maximum relative error of 38.3%. Increasing the assumed frontal obstruction height from 1.030 to 1.160 m increased the estimated blind-spot distance from 8.583 to 21.091 m, demonstrating substantial sensitivity to visibility assumptions. In an exploratory spatial validation of 27 vehicles, all three assessors classified 24 cases (88.9%) as Valid or Partially Valid, although agreement on individual four-category verdicts was moderate (Fleiss’ κ=0.512). These qualitative results describe the evaluated sample and do not establish population-level performance or physical blind-spot accuracy. The findings support the feasibility of using GT 3D bounding boxes as a compact geometric representation for vehicle-level blind-spot analysis while highlighting the need for vehicle-specific visibility parameters and independent physical validation.
Source: Sensors
DOI: 10.3390/s26217093
Authors: Heeju Je, Seoyoung Choi, Beom Kwon
Institution: Dongduk Women's University
Language: English
Topic: Traffic and Road Safety
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $3,137
OpenAlex ID: W7225656432
FWCI:
Cites:
Cited by:
Related to: