Artificial intelligence’s rapid advancement has enabled various applications, including intelligent video surveillance systems, assisted living, and human–computer interaction. These applications often require one core task: video-based human action recognition. Research in human video-based human action recognition is vast and ongoing, making it difficult to assess the full scope of available methods and current trends. This survey concisely explores the vision-based human action recognition field and defines core concepts, including definitions and explanations of the common challenges and most used datasets. Additionally, we provide in an easy-to-understand manner the literature approaches and their evolution over time, emphasizing intuitive notions. Finally, we explore current research directions and potential future paths. The core goal of this work is to provide future works with a shared understanding of fundamental ideas and clear intuitions about current works and find new research opportunities.

Training a model to recognize human actions in videos is computationally intensive. While modern strategies employ transfer learning methods to make the process more efficient, they still face challenges regarding flexibility and efficiency. Existing solutions are limited in functionality and rely heavily on pretrained architectures, which can restrict their applicability to diverse scenarios. Our work explores knowledge distillation (KD) for enhancing the training of self-supervised video models in three aspects: improving classification accuracy, accelerating model convergence, and increasing model flexibility under regular and limited-data scenarios. We tested our method on the UCF101 dataset using differently balanced proportions: 100%, 50%, 25%, and 2%. We found that using knowledge distillation to guide the model’s training outperforms traditional training without affecting the classification accuracy and while reducing the convergence rate of model training in standard settings and a data-scarce environment. Additionally, knowledge distillation enables cross-architecture flexibility, allowing model customization for various applications: from resource-limited to high-performance scenarios.

Artificial intelligence’s rapid advancement has enabled various applications, including intelligent video surveillance systems, assisted living, and human-computer interaction. These applications often require one core task: video-based human action recognition. Research in human video-based human action recognition is vast and ongoing, making it difficult to assess the full scope of available methods and current trends. This survey provides an in-depth exploration of the vision-based human action recognition field, comprehensively offering the available techniques and their evolution, highlighting the cutting-edge ideas driving its development. We also analyze the most used keywords in research papers over the past years to identify trends and predict possible future directions. Hence, this concise survey helps researchers understand the breadth of existing approaches, evaluate current research trends, and stay up-to-date on potential developments.

Human action recognition is an active field of research that intends to explain what a subject is doing in an input video. Deep learning architectures serve as the foundation for cutting-edge approaches. Recent research, on the other hand, indicates that hand-crafted characteristics are complementary and, when combined, can enhance classification accuracy. Cutting-edge approaches are based on deep learning architectures. Recent research, however, indicates that hand-crafted features complement each other and can help boost classification accuracy when combined. We introduce the key trajectories approach that is based on the popular, hand-crafted method, improved dense trajectories. Our work explores how pose estimation can be used to find meaningful key points to reduce computational time, undesired noise, and to guarantee a stable frame processing rate. Furthermore, we tested how feature-tracking behaves with dense inverse search and with a frame to frame subject key point estimation. Our proposal was tested on the KTH and UCF11 datasets employing Bag-of-words and on the UCF50 and HMDB datasets using Fisher Vector, where we got an accuracy performance of 95.71, 84.88, 92.9, and 81.3%, respectively. Also, our proposal can recognize subject actions in video eight times faster compared to its dense counterpart. To maximize the bag-of-words classification performance, we illustrate how the hyperparameters affect both accuracy and computation time. Precisely, we present an exploration of the vocabulary size, the SVM hyperparameter, the descriptor’s distinctiveness, and the subject body key points.