Optim3D: Efficient and Scalable Generation of Large-Scale 3D Building Models

3D building reconstruction in the urban environment is a key step in developing urban digital twins. So far, the existing researches focus on the modeling process and overlook the scalability issues related to the extensive data and iterative calculations. In this paper, we propose an approach for efficient and scalable 3D generation of large-scale building models from aerial LiDAR data. Our method involves data indexing and tiling for enhanced efficiency, along with parallel processing to accelerate the modeling process. We also correct ground-floor elevation inaccuracies in the final model to ensure reliable applications and analysis. The results show that, for large datasets, the indexing overcomes the memory saturation and speeds up the tiling and the reconstruction processes. The parallel processing accelerates the reconstruction process by nearly five times compared to an iterative approach. Additionally, using K-nearest neighbors instead of radius search resulted in more accurate elevation values. Our source code is available at: https://github.com/Yarroudh/Optim3D.