A Multi-Stage Pipeline for 3D Reconstruction of Vehicles from Heterogeneous Multi-View Images

Background/Aim: Three-dimensional (3D) reconstruction of vehicles remains a challenging task, particularly when aiming to generate standardized models from multiple samples of the same make, model, body type, and year. While traditional Structure-from-Motion approaches excel at reconstructing individual objects, they struggle to create generalized models from diverse multi-view datasets exhibiting variations in lighting, background, and appearance. This study aims to overcome these limitations by presenting a novel automated pipeline for standardized 3D vehicle reconstruction.Methods: The proposed methodology employs a three-stage approach: (1) preprocessing with background removal and adaptive histogram equalization to isolate vehicle regions and normalize illumination; (2) deep learning-based orientation classification utilizing an EfficientNet-B0 architecture to categorize images into eight directional views; and (3) sparse reconstruction via hierarchical localization and COLMAP, implementing a neighbor-based image pairing strategy.Results: The developed pipeline was applied to a diverse collection of heterogeneous, multi-source vehicle images. This process resulted in the successful generation of 3D Gaussian Splatting models for over a hundred distinct vehicle classes.Conclusion: This study provides a robust framework for scalable 3D vehicle modeling. To support reproducibility and facilitate future research in generative AI and autonomous systems, the resulting dataset will be made publicly available. This contribution establishes a standardized benchmark resource for tasks such as novel view synthesis, damage assessment, and vehicle appearance modeling, representing one of the first publicly available datasets of vehicle-class-specific 3D Gaussian Splatting reconstructions derived from heterogeneous multi-source collections.