Domain Adaptation for Human Activity Recognition

Human activity recognition (HAR) involves classifying and identifying human activities based on sensor data, with applications in healthcare, sports, and warehouses. Accurate data analysis can significantly benefit these domains by improving efficiency and safety.

We are developing a domain adaptation algorithm for HAR, where a machine/deep learning model trained on one dataset (source) is applied to a different dataset (target) with distinct characteristics. Our focus is on using a self-training approach to reduce reliance on manual annotation while enhancing the generalizability of machine/deep learning models across different domains and environments.

Additionally, we are integrating state-of-the-art methods such as transformers into our algorithm to achieve a highly effective solution for HAR. This integration aims to improve the model's performance and adaptability to various real-world applications.

Key Publications

Moh'd Khier Al Kfari, Stefan Lüdtke. Domain Adaptation in Human Activity Recognition through Self-Training. Companion of the 2024 on ACM International Joint Conference on Pervasive and Ubiquitous Computing (Ubicomp Workshops) 2024. [web]