INTELLIGENT RECOGNITION OF MULTIMODAL HUMAN ACTIVITIES FOR PERSONAL HEALTHCARE

Abstract

The advancement of wearable sensors, smart devices, and artificial intelligence has enabled the development of intelligent healthcare systems capable of monitoring human activities in real time. Human Activity Recognition (HAR) plays a crucial role in personal healthcare by analyzing daily activities such as walking, sitting, running, and sleeping to assess an individual’s physical condition and detect potential health risks. Traditional activity recognition systems often rely on single-modal data, which limits their accuracy and robustness in real-world scenarios. To overcome these limitations, this project proposes an intelligent multimodal human activity recognition system that integrates multiple data sources for improved performance and reliability. The proposed system utilizes data from various modalities, including wearable sensors (accelerometer, gyroscope), video data, and physiological signals such as heart rate. Data preprocessing techniques such as noise filtering, normalization, and segmentation are applied to ensure data quality. Feature extraction methods are used to capture temporal, spatial, and statistical characteristics of human activities. Machine learning algorithms such as Random Forest, Support Vector Machine (SVM), and k-Nearest Neighbors (k-NN) are employed for classification, while deep learning models such as Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks are used to capture complex patterns and sequential dependencies in the data. The system is evaluated using performance metrics such as accuracy, precision, recall, and F1-score, demonstrating improved performance compared to single-modal approaches. The integration of multimodal data enhances activity recognition accuracy and provides more reliable health monitoring.