2D outlining of buildings using airborne point cloud data and deep learning
2D outlines or footprints of buildings are either generated from high resolution images or point cloud data. Often, it consists of a sequence of processing steps, such as edge detection, contouring, generalization and regularization to achieve well-shaped building outlines. Central challenge in this MSc research is to analyze whether you can find or design a deep learning network that can learn from existing 2D building footprints and high density airborne point cloud data. Optional, it can be decided to extent this work to 3D building model generation using deep learning.
Literature review on 2D outlining of buildings, and delineating techniques using deep learning. Implement suitable deep learning network, generate training data and 2D outlines on dense point clouds of buildings. Analyze 2D outlines, finetune training data and network parameters, perform accuracy assessment.