ActiSense AI: Real-Time Human Activity Recognition Using Wearable Sensor Intelligence

Abstract

This paper presents ActiSense AI, a real-time human activity recognition (HAR) system leveraging wearable inertial measurement units (IMUs) coupled with a hybrid deep learning architecture. The proposed system integrates accelerometer, gyroscope, and heart rate sensors embedded in a wristband form factor with an edgedeployed Convolutional Bidirectional Long Short-Term Memory (CNN-BiLSTM) model. ActiSense AI recognizes eight distinct daily activities including walking, running, sitting, standing, stair climbing, lying, cycling, and jumping with real-time inference latency below 120 milliseconds. Evaluated on the UCI-HAR, PAMAP2, and a custom dataset, the proposed system achieves 96.7% classification accuracy, outperforming conventional machine learning and deep learning baselines. The system demonstrates practical viability for healthcare monitoring, elderly care, sports analytics, and rehabilitation applications.