Real-Time Human Activity Recognition in Smart Homes Using IoT Sensors and Deep Learning Models
This study answers the challenge of real-time human activity identification in smart homes. Further, the authors suggest an end-to-end solution using IoT, deep learning, smart homes, activity recognition, real-time sensor fusion, and model evaluation. The purpose of this project is to establish a solid framework that combines different sensor signals in real-time with a hybrid CNN-LSTM model for detecting as well as classifying human actions. New ways of getting data ready and extracting anything useful were employed for minimizing noise in sensors and model training. Through experimental evaluation, the proposed method obtains 92.5% accuracy, 91.0% precision, 90.2% recall, and 90.6% F1-score and greatly exceeds the current methods. The system also has an average latency of 150 ms, hence suited for real-time applications. These results suggest that, by merging modern deep learning architectures with effective sensor fusion, the performance of the activity recognition system may be considerably boosted. In addition, it aids in increasing the safety, energy management, and enjoyment in smart homes. The proposed method doesn't merely improve existing procedures but also provides a strong platform for future innovations aimed towards smart home technologies.
Paper
Full text
Real-Time Human Activity Recognition in Smart Homes Using IoT Sensors and Deep Learning Models
Semantic Scholar · 2025
Abstract
This study answers the challenge of real-time human activity identification in smart homes. Further, the authors suggest an end-to-end solution using IoT, deep learning, smart homes, activity recognition, real-time sensor fusion, and model evaluation. The purpose of this project is to establish a solid framework that combines different sensor signals in real-time with a hybrid CNN-LSTM model for detecting as well as classifying human actions. New ways of getting data ready and extracting anything useful were employed for minimizing noise in sensors and model training. Through experimental evaluation, the proposed method obtains 92.5% accuracy, 91.0% precision, 90.2% recall, and 90.6% F1-score and greatly exceeds the current methods. The system also has an average latency of 150 ms, hence suited for real-time applications. These results suggest that, by merging modern deep learning architectures with effective sensor fusion, the performance of the activity recognition system may be considerably boosted. In addition, it aids in increasing the safety, energy management, and enjoyment in smart homes. The proposed method doesn't merely improve existing procedures but also provides a strong platform for future innovations aimed towards smart home technologies.