Reconstructing 3D model of accident scene using drone image processing

Norahim, Mohamad Norsyafiq Iman; Tahar, Khairul Nizam; Maharjan, Gyanu Raja; Matos, Jose C. · 2023 · Crossref

DOI: 10.11591/ijece.v13i4.pp4087-4100

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 inefficiencies and inaccuracies inherent in traditional accident scene reconstruction methods, which rely on manual sketching, tape measurements, and 2D photography. These conventional techniques are time-consuming, prone to human error, and often result in prolonged traffic congestion and insufficient evidence for legal proceedings. The research aims to evaluate the effectiveness of using Unmanned Aerial Vehicles (UAVs) to reconstruct 3D models of accident scenes, offering a faster, more precise, and cost-effective alternative to terrestrial laser scanning and manual surveys. The methodology involved a four-phase process: preliminary work, data acquisition, data processing, and data analysis. The study was conducted at a simulated accident site at Universiti Teknologi MARa, Shah Alam, Malaysia. Researchers utilized a DJI drone to capture imagery under five distinct flight configurations: circular missions at 5, 7, and 10 meters altitude, and double-grid and single-grid missions at 10 meters altitude. Ground Control Points (GCPs) were established using GNSS with Real-Time Kinematic (RTK) technology to ensure georeferencing accuracy. Data processing was performed using Pix4D software to generate 3D textured meshes, point clouds, and orthomosaics. Accuracy was assessed by comparing measurements derived from the 3D models against actual site measurements using the Root Mean Square Error (RMSE). The results demonstrated that all tested flight parameters could produce usable 3D models, but accuracy varied significantly based on flight pattern, altitude, and the inclusion of GCPs. The circular flight method at 5 meters altitude with GCPs yielded the highest accuracy, achieving an RMSE of 0.047 meters. This method also produced the densest point cloud (8,941,091 points) and the clearest mesh visualization. Without GCPs, the same configuration resulted in an RMSE of 0.067 meters, highlighting the necessity of ground control for precision. Higher altitudes (7 and 10 meters) and grid-based missions resulted in higher RMSE values (ranging from 0.061 to 0.086 meters) and inferior visual quality in the point clouds and meshes, particularly for the single-grid method. The study concludes that UAV-based photogrammetry is a viable and superior replacement for traditional site measurement techniques in accident reconstruction. The circular flight pattern at low altitude (5 meters) combined with GCPs is identified as the optimal approach for achieving high-precision 3D reconstructions. This method allows investigators to rapidly capture comprehensive scene data, reduce traffic disruption, and provide detailed, measurable 3D evidence that can be re-analyzed post-incident, thereby improving the reliability of forensic investigations and legal outcomes.

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.

• in depth crash investigation