Dynamically generated landscapes

Abstract

Artificial Intelligence (AI) has allowed different fields within technology to flourish, however, the area of terrain generation has not been able to experience this same growth. This area makes use of highly specialised software to create virtual environments, but this type of software usually require their users to be experienced in this field. In this paper, we investigate different approaches that can be used for the purposes of terrain generation and gesture recognition. For the purpose of terrain generation we look at the use of Generative Adversarial Networks (GANs), Diffusion Models, and Noise Generators. For each if theses approaches we created a system that made use of it, which were then compared against one another. A gesture recognition system was then created. This gesture recognition system makes use of body tracking algorithms, hand gesture recognition models and depth cameras. This system allowed us to track a users body in 3D-space by using the X and Y co-ordinates of the users body parts within the 2D image in conjunction with the depth sensor values. A combined system that makes use of noise generation algorithms and the gesture recognition system was created. This system was able to carry out terrain generation using the stochastic inputs provided by the user in the way of gestures, where each gesture carried out it’s own assigned process. It was concluded that while terrain generation approaches that make use of AI result in good results, they have yet to reach the quality that can be seen when using specialised terrain generation software which make use of noise generators.