Back to research

IEEE Data Descriptions / Scholar indexed / Mendeley Data

RGB Video of Natural Hand Gestures Dataset (RGB-NHG)

A public smartphone RGB gesture dataset and descriptor publication for RGB gesture recognition with consumer-grade camera data.

Dataset Goal

The dataset supports gesture-recognition research using ordinary smartphone RGB video, without depth or infrared sensors.

Descriptor Publication

The Scholar-indexed descriptor is listed as MT Rahat, D. Sarker, and K. Nur, "Descriptor: RGB Video of Natural Hand Gestures Dataset (RGB-NHG)," IEEE Data Descriptions, pp. 1-10.

Collection Design

It includes 13 everyday hand gestures captured with a Samsung Galaxy S23 under bright daylight and dim green LED lighting, with both hand-only and full-body views.

Benchmark Setup

The dataset contains 312 standardized video clips, processed at 640 by 640 pixels and 30 fps, with extracted frame sequences and a labels.csv metadata file.

Research Use

It is designed for robustness studies involving lighting variation, user diversity, mobile deployment, gesture metadata, and comparisons between hand-only and full-body recognition.

What Was Difficult

Practical gesture datasets must balance realism, repeatability, capture variation, labeling consistency, and privacy-aware organization.

Open Directions

  • Can full-body context improve recognition when hand-only signals are ambiguous?
  • Which augmentations preserve gesture identity across lighting conditions?
  • How should benchmark splits handle subject and capture-mode variation?