Next Generation Human Action Recognition: A Comprehensive Review of State-of-the-Art Signal Processing Techniques

Abstract

Human Action Recognition is a rapidly evolving field at the intersection of computational intelligence, signal processing, and machine learning, driving innovations in healthcare, fitness, surveillance, and smart environments. This review bridges the gap between classical techniques, such as Fourier Transform (FT), Wavelet Transform (WT), and Principal Component Analysis (PCA), and advanced Deep Learning (DL) approaches, including Convolutional Neural Network (CNN), Recurrent Neural Network (RNN) and emerging Transformer-based architectures. A key focus is the integration of multimodal sensing modalities, including accelerometers, gyroscopes, Electromyography (EMG), and electroencephalography (EEG), in conjunction with noncontact technologies such as thermal and infrared imaging, enhancing HAR adaptability in diverse environments. Unlike prior surveys, this work presents a comprehensive synthesis of HAR methodologies, systematically analyzing 217 studies to quantify trends, assess challenges such as noise resilience, real-time constraints, and scalability, and highlight innovative solutions, including advanced multimodal fusion frameworks and energy-efficient architectures. Furthermore, a research roadmap is proposed to advance scalable, real-time HAR systems with deeper multimodal data integration and lightweight computational models, paving the way for robust and adaptable solutions in real-world applications.