A Comprehensive Survey and Analysis of CNN-LSTM-Based Approaches for Human Activity Recognition

In recent years, the field of Human Activity Recognition (HAR) has become increasingly vital in the realm of computer vision, unveiling a plethora of practical applications that extend far and wide. From enhancing surveillance systems to enabling precise activity tracking, delving into sports analysis, and facilitating efficient event identification, HAR has emerged as a transformative technology. Machine Learning, as a driving force in this arena, has ushered in a new era of methodologies, each adding its own layer of sophistication to HAR systems. Convolutional Neural Networks (CNNs) bring in the power of visual hierarchy and feature learning, while Graph Neural Networks excel in capturing complex relationships within activity data. Long Short-Term Memory (LSTM) networks, with their ability to capture temporal dependencies, have further fortified the capabilities of HAR models. This comprehensive article endeavors to provide an extensive overview of the various categories within HAR and the nuanced methodologies employed across these categories. By delving into the intricacies of each approach, we aim to offer a nuanced understanding for researchers engaged in HAR. Beyond a mere compilation of facts, this survey article seeks to be a guiding light, aiding researchers in the discovery of pertinent references and serving as a valuable resource for shaping the trajectory of future research in the dynamic field of Human Activity Recognition.