Real-Time Human Activity Recognition Using Edge Artificial Intelligence

Abstract

Human Activity Recognition (HAR) has become one of the most important research areas in artificial intelligence because of its extensive applications in healthcare monitoring, smart homes, industrial safety, surveillance, rehabilitation, fitness tracking, and human–computer interaction. Conventional cloud-based HAR systems collect sensor or video data and transmit them to centralized servers for processing. Although cloud computing offers high computational capability, it introduces communication latency, bandwidth consumption, network dependency, and privacy concerns that limit its applicability in real-time scenarios. The emergence of Edge Artificial Intelligence (Edge AI) enables intelligent inference directly on edge devices such as smartphones, embedded processors, wearable sensors, and IoT gateways, thereby reducing response time while preserving user privacy. Substantial advances were achieved in lightweight deep learning models, convolutional neural networks (CNNs), recurrent neural networks (RNNs), Long Short-Term Memory (LSTM) networks, TinyML, TensorFlow Lite, model compression, and edge computing frameworks for human activity recognition. These technologies significantly improved recognition accuracy while satisfying the computational constraints of resource-limited edge devices.