A Survey of Deep Learning-based Hand Pose Estimation

Abstract

Hand pose estimation is one of the representative tasks in computer vision. Solving the hand pose estimation problem is essential for various fields such as virtual reality, augmented reality, mixed reality, and human-computer interaction. Due to the significant development of deep learning techniques, the hand pose estimation task has reached significant performance on many hand pose estimation datasets. However, the hand pose estimation task still faces many challenges due to the lack of large-scale labeled data, severe occlusion, low hand resolution, and background clutter. To better understand the hand pose estimation task, this paper presents a comprehensive survey of outstanding papers over the last five years. The paper first introduces 19 common hand pose estimation datasets, then extensively discusses some of the mainstream approaches in hand pose estimation, including fully supervised, semi-supervised, weakly supervised, and self-supervised learning methods. Finally, we extensively discuss the future evolution trends of hand pose estimation.