Attention at Hand: A Comprehensive Survey of Transformer-based Hand Gesture and Landmark Detection Models

Transformer-based architectures, originally developed for natural language processing, have recently shown significant promise in the field of computer vision, particularly for hand gesture recognition and landmark detection. These models utilize attention mechanisms to efficiently capture spatial and temporal dependencies, making them well-suited for complex tasks involving intricate hand movements. This survey explores the evolution and current state-of-the-art in transformer-driven approaches, including pure transformer models, hybrid systems combining CNNs or GCNs, and task-specific adaptations tailored for gesture interpretation. We analyze their effectiveness in addressing challenges such as occlusion, lighting variability, and real-time processing demands. Additionally, we highlight key application domains like virtual reality, sign language translation, and human-computer interaction. The paper concludes with a discussion on current limitations, such as computational overhead and data scarcity, and outlines future directions for research and deployment in real-time systems.