In this webinar, we demonstrate how to prepare a training and validation dataset for applying weakly-supervised deep learning models using LIDAR data from Glasgow as our example.

City 3D models can be a useful data source not just for 3D visualization, but also for urban environment assessment and analysis such as flooding simulation, urban housing planning application, and urban climate modeling. At the Urban Big Data Centre, we use high-density airborne LiDAR point clouds to extract building and tree 3D information on a city-wide scale.

However, LiDAR point cloud processing and high-quality 3D model construction are demanding in terms of time and labour. In this webinar, we will share our experience of how to extract building and tree 3D information efficiently from the raw high-density airborne LiDAR point clouds. Methods of preparing a training and validation dataset, and training a weakly-supervised deep learning model to segment target objects (i.e., buildings and trees) will be introduced. A set of Glasgow LiDAR data will be used as an example to demonstrate the workflow.

This webinar is ideal for researchers, students, data scientists, or anybody interested in LiDAR remote sensing or using 3D city models for their research or work.