Radio Signal Strength Indication Augmentation for One-Shot Learning in Indoor Localisation

Collecting annotated Radio Signal Strength Indication (RSSI) data from wearables for indoor localisation is a time consuming and laborious process. For healthcare related applications, where indoor localisation is seeing increasing use, it is often more difficult due to limitations surrounding a patient's condition. In this work we propose simple yet effective RSSI augmentation techniques for indoor localisation which boosts performance in scenarios where there is as little as a single label. Specifically, we propose the use of two augmentations to the signal, reflective of common variations in real settings. In order to validate their effectiveness, we frame the problem as one-shot learning and use a dataset collected in several typical residential homes involving realistic in-home behaviours. We show that our proposed augmentations can increase indoor localisation accuracy in four homes by up to 22.94 percentage points with only one short training sample collected in each room.