BISINDO DATASET
Description
This dataset contains hand gesture images representing the 26 letters (A–Z) of the Indonesian Sign Language (BISINDO) alphabet. The gestures were performed according to the official BISINDO alphabet reference published by the Indonesian Ministry of Primary and Secondary Education (Kementerian Pendidikan Dasar dan Menengah Republik Indonesia). The dataset was developed to support research in computer vision, hand landmark detection, gesture recognition, and machine learning applications for Indonesian Sign Language translation. Each class corresponds to one alphabet letter (A–Z). The dataset can be used for image classification, hand landmark extraction using MediaPipe Hands, feature engineering, and the development of machine learning models for BISINDO alphabet recognition. It is intended for academic research, educational purposes, and benchmarking of sign language recognition systems.
Files
Steps to reproduce
The dataset was collected by capturing hand gesture images representing the 26 letters (A–Z) of the Indonesian Sign Language (BISINDO) alphabet. All gestures were performed according to the official BISINDO alphabet reference published by the Indonesian Ministry of Primary and Secondary Education (Kementerian Pendidikan Dasar dan Menengah Republik Indonesia). Image acquisition was carried out using an ELP 180° fisheye USB camera connected to a computer. During data collection, participants positioned their hands at an approximate distance of 30–60 cm from the camera lens. Images were recorded under various natural hand orientations and slight pose variations to increase the diversity of the dataset while maintaining the correct BISINDO hand gesture for each alphabet letter. A total of seven participants were involved in the data collection process using a random sampling approach. Each participant performed all 26 BISINDO alphabet gestures, and multiple images were captured for every gesture to account for natural variations in hand shape, orientation, and positioning. The collected images were organized into separate folders corresponding to each alphabet class (A–Z). This dataset can be reproduced by following the same acquisition protocol: using an ELP 180° fisheye camera, maintaining a hand-to-camera distance of approximately 30–60 cm, following the official BISINDO alphabet reference, collecting data from multiple participants, and storing the captured images according to their respective alphabet labels.
Institutions
- State University of MalangEast Java, Malang