Digital Reconstitution of Road Traffic Accidents: A Flexible Methodology Relying on UAV Surveying and Complementary Strategies to Support Multiple Scenarios

An Approach to the Automatic Construction of a Road Accident Scheme Using UAV and Deep Learning Methods

Recreating a road traffic accident scheme is a task of current importance. There are several main problems when drawing up a plan of accident: a long-term collection of all information about an accident, inaccuracies, and errors during manual data fixation. All these disadvantages affect further decision-making during a detailed analysis of an accident. The purpose of this work is to automate the entire process of operational reconstruction of an accident site to ensure high accuracy of measuring the distances of the relative location of objects on the sites. First the operator marks the area of a road accident and the UAV scans and collects data on this area. We constructed a three-dimensional scene of an accident. Then, on the three-dimensional scene, objects of interest are segmented using a deep learning model SWideRNet with Axial Attention. Based on the marked-up data and image Transformation method, a two-dimensional road accident scheme is constructed. The scheme contains the relative location of segmented objects between which the distance is calculated. We used the Intersection over Union (IoU) metric to assess the accuracy of the segmentation of the reconstructed objects. We used the Mean Absolute Error to evaluate the accuracy of automatic distance measurement. The obtained distance error values are small (0.142 ± 0.023 m), with relatively high results for the reconstructed objects’ segmentation (IoU = 0.771 in average). Therefore, it makes it possible to judge the effectiveness of the proposed approach.

Accuracy Assessment of Small Unmanned Aerial Vehicle for Traffic Accident Photogrammetry in the Extreme Operating Conditions of Kuwait

This study presents the first accuracy assessment of a low cost small unmanned aerial vehicle (sUAV) in reconstructing three dimensional (3D) models of traffic accidents at extreme operating environments. To date, previous studies have focused on the feasibility of adopting sUAVs in traffic accidents photogrammetry applications as well as the accuracy at normal operating conditions. In this study, 3D models of simulated accident scenes were reconstructed using a low-cost sUAV and cloud-based photogrammetry platform. Several experiments were carried out to evaluate the measurements accuracy at different flight altitudes during high temperature, low light, scattered rain and dusty high wind environments. Quantitative analyses are presented to highlight the precision range of the reconstructed traffic accident 3D model. Reported results range from highly accurate to fairly accurate represented by the root mean squared error (RMSE) range between 0.97 and 4.66 and a mean percentage absolute error (MAPE) between 1.03% and 20.2% at normal and extreme operating conditions, respectively. The findings offer an insight into the robustness and generalizability of UAV-based photogrammetry method for traffic accidents at extreme environments.

Accuracy of 3D Landscape Reconstruction without Ground Control Points Using Different UAS Platforms

The rapid increase of low-cost consumer-grade to enterprise-level unmanned aerial systems (UASs) has resulted in the exponential use of these systems in many applications. Structure from motion with multiview stereo (SfM-MVS) photogrammetry is now the baseline for the development of orthoimages and 3D surfaces (e.g., digital elevation models). The horizontal and vertical positional accuracies (x, y and z) of these products in general, rely heavily on the use of ground control points (GCPs). However, for many applications, the use of GCPs is not possible. Here we tested 14 UASs to assess the positional and within-model accuracy of SfM-MVS reconstructions of low-relief landscapes without GCPs ranging from consumer to enterprise-grade vertical takeoff and landing (VTOL) platforms. We found that high positional accuracy is not necessarily related to the platform cost or grade, rather the most important aspect is the use of post-processing kinetic (PPK) or real-time kinetic (RTK) solutions for geotagging the photographs. SfM-MVS products generated from UAS with onboard geotagging, regardless of grade, results in greater positional accuracies and lower within-model errors. We conclude that where repeatability and adherence to a high level of accuracy are needed, only RTK and PPK systems should be used without GCPs.