Reproduce “A Survey of Human Activity Recognition in Smart Homes Based on IoT Sensors Algorithms: Taxonomies, Challenges, and Opportunities with Deep Learning” on OpenResearch

About this paper

Damien Bouchabou, Sao Mai Nguyen, Christophe Lohr, Benoit Leduc, Ioannis Kanellos

A survey evaluates Human Activity Recognition (HAR) in smart homes through IoT ambient sensors, proposing a new challenge-centric taxonomy and reviewing current algorithms, particularly deep learning techniques. It identifies persistent issues like data variability and the need for standardized evaluation, while outlining future research directions.

Key insights

• Deep learning models, particularly CNNs and LSTMs, effectively handle automatic feature extraction and temporal dependencies in sparse and irregular ambient sensor data, outperforming traditional machine learning.
• Significant challenges persist in recognizing complex, interleaved, or multi-user activities, as well as in adapting to temporal drift and variability across different smart home settings.
• The field is hampered by a critical lack of standardized, representative datasets and uniform evaluation protocols, which impedes objective comparison and robust research progress.

Summary by alphaXiv.