3D image reconstruction after traffic accident

• Autor(a) principal: Valério, Pedro Miguel Soares
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/35183
• topic: Photogrammetry; 3D reconstruction; Drone; Traffic accidents; Point cloud; 3D model

In the investigation of a road traffic accident, it is crucial to obtain a sufficiently detailed information model of the scene. In recent years, the study of digital 3D reconstruction has seen great progress and has been applied to many fields, for its effectiveness in correctly capturing realworld data. The process involves mapping a physical space to a set of points in a digital space, forming a point cloud mesh. This mesh is then processed and transformed into other 3D products, such as triangular models and textures. There are multiple approaches to obtain the point cloud data, most notably, based on laser-scanning, like LiDAR. While these techniques yield very accurate results, they are typically very expensive, lengthy and difficult to use. In traffic accident investigation, it is necessary, not only to obtain an accurate model of the scene, but also to do so as quickly as possible, in order to minimise the disruption of regular traffic. Photogrammtery is a 3D reconstruction technique which obtains spatial information from photographic images. A recent application of photogrammetry uses drones to acquire these images. Drones can be used to fly over an area and capture images from a wide variety of vantage points and different angles. Combined with their relatively low cost and great flexibility, this makes them ideal for data collection and monitoring terrain with difficult access. Moreover, a drone can survey a large area very quickly, without the need to set up control stations as is the case with most LiDAR approaches. This work aims to analyse the process of 3D reconstruction from a set images of real traffic accidents acquired by drone. The various stages of generating 3D products, such as point clouds and triangular meshes are studied, in order to obtain a reliable representation of the accident scenes. For this purpose, commercial and other presently used software and algorithms are tested and compared in the generation of these 3D representations as well as which factors and generation parameters have the greatest influence in the quality of the final products.