Euclidean consistency-driven dual-layer information fusion framework for UAV-based traffic accident scene reconstruction

Abstract

This study aims to develop an interpretable dual-layer information fusion framework driven by the Euclidean Consistency Index (ECI) for UAV flight parameter optimization, which integrates a statistically interpretable linear module with a nonlinear learning layer implemented via support vector regression to enhance 3D reconstruction accuracy in traffic accident scenarios. Experiments based on UAV oblique photogrammetry (27 configurations) were conducted to evaluate reconstruction accuracy using elevation error, horizontal error, and distortion anomaly metrics. The results reveal that flight altitude and image overlap exert coupled effects on reconstruction performance. An optimal parameter range is identified as 20–25 m altitude, 80–85% forward overlap, and 70–75% side overlap, achieving centimeter-level accuracy and stable geometric consistency. The high inter-layer agreement (ECI = 0.9133) confirms the robustness of the proposed fusion strategy. Validation in a real-world accident scene offers preliminary support for its practical applicability under similar conditions. Within the tested experimental scope, this framework provides a quantitatively interpretable and operationally stable foundation for UAV flight parameter selection in accident reconstruction.