Environment Capture and Simulation for UAV Cinematography Planning and Training

This paper presents a workflow for the generation of environmental models, which can be employed for training and planning of drone based shooting. This converts multiple 2D environmental and terrain images into 3D models and height maps, which are then imported into a simulation engine as environment assets. A subjective study has also been conducted to characterise the relationship between reconstruction quality and the number and location of input images for various environmental scenarios. Using this workflow, demonstration videos have been produced which combine extracted environments with an existing object model (cyclist) in simulation. These illustrate its utility in shot planning, rehearsal and training.