A Smartphone Camera Based RGB Video Dataset of Natural Hand Gestures

This dataset provides RGB video recordings and corresponding frame sequences for 13 everyday hand gestures, captured using a Samsung Galaxy S23 smartphone under real-world lighting conditions. Designed for research on gesture recognition using only consumer-grade cameras, the dataset excludes depth or infrared sensors to emphasize feasibility with standard hardware.

Each gesture was performed by two individuals, across three takes, under both bright daylight and dim green LED lighting. To capture different spatial contexts, recordings were made in two views: hand-only (focused solely on the hand) and full-body (showing the entire participant). This results in a total of 312 standardized video clips: 13 gestures × 2 participants × 3 takes × 2 lighting conditions × 2 capture modes.

Each video was processed to a uniform resolution of 640×640 pixels at 30 fps, and frames were extracted into organized image sequences. The data is arranged by gesture class and capture mode, with a labels.csv file at the root providing metadata for each clip. This includes gesture label, participant ID, lighting condition, view type, and take number, along with the corresponding frame path.

The dataset is ideal for training and evaluating machine learning models for gesture recognition, particularly in scenarios that require robustness to lighting variations, user diversity, and mobile deployment.