MUMBAI, India, Feb. 6 -- Intellectual Property India has published a patent application (202641007299 A) filed by Sahana M. Kulkarni; and Dr. Jamuna S, Bengaluru, Karnataka, on Jan. 24, for 'human activity recognition from multi-sensor data via unified feature engineering and optimized deep learning.'
Inventor(s) include Sahana M. Kulkarni; and Dr. Jamuna S.
The application for the patent was published on Feb. 6, under issue no. 06/2026.
According to the abstract released by the Intellectual Property India: "A system and method for Human Activity Recognition (HAR) using multi-sensor wearable data is disclosed. The invention addresses challenges arising from sensor noise, missing data, high dimensionality, and class imbalance that limit the practical deployment of existing deep learning-based HAR systems. The proposed framework integrates composite feature engineering with mutual information-based feature selection and a computationally efficient, optimized Multi-Layer Perceptron (MLP). Composite features, including mean body temperature and acceleration magnitude, are generated to enhance discriminative capability while reducing redundancy. Informative features are selected using mutual information to eliminate irrelevant signals and improve robustness. The selected features are classified using an MLP employing Swish activation, batch normalization, dropout, and label smoothing to enhance generalization under noisy and imbalanced conditions. Experimental validation demonstrates an activity recognition accuracy exceeding 98%, with significantly reduced computational complexity compared to convolutional and recurrent neural networks. The invention is scalable, interpretable, and suitable for real-time deployment on wearable, embedded, and IoT-based healthcare systems."
Disclaimer: Curated by HT Syndication.
