Subject

Neural networks (NNs) are fundamental for nearly every computer vision task, such as image classification and human pose estimation. The integration of NNs in Augmented Reality (AR) systems, which overlay virtual 3D objects onto the real world, would therefore be beneficial for a wide range of applications. This integration is usually performed by running the NNs on an external server, which receives the input data from the AR headset (HoloLens) and sends the results back via Wi-Fi. This approach has several limitations, as the Wi-Fi introduces latency and is unstable.

Kind of work

The objective of this master thesis is to investigate which NNs architectures can run directly on HoloLens 2, without using an external server connected via Wi-Fi. Moreover, different approaches to run inference directly on HoloLens 2 will be evaluated.