Markerless RGB Dataset of Hand Joint Kinematics Across Five Grasp Types

This dataset presents a markerless RGB-based collection of continuous 3D hand joint kinematics across five standardized object-interaction grasp types. Data was acquired using an RGB camera and MediaPipe Hands, providing an accessible alternative to marker-based motion capture systems for biomechanical analysis. The dataset includes recordings from 10 healthy subjects (6 male, 4 female), comprising approximately 28,000 validated active-hold frames. Each frame captures 11 key anatomical joint angles, including thumb (CMC, MCP, IP) and finger (MCP, PIP) joints, ensuring robust and reliable kinematic representation. Participants performed five standardized grasp types using everyday objects: power grasp (bottle), tripod grip (pen), static hold (smartphone), precision pinch (coin/paper), and lateral pinch (book spine). Each grasp was recorded across multiple trials under controlled conditions. The dataset is organized into:- Raw continuous recordings (including rest and transitions)- Processed active-hold frames for analysis- Visualization outputs and statistical summaries- Source code for data capture and processing This dataset supports applications in hand pose estimation, human-object interaction analysis, biomechanics, rehabilitation research, and machine learning. All data are fully anonymized and contain no personally identifiable information. License:- Dataset: Creative Commons Attribution 4.0 (CC BY 4.0)- Code: MIT License