Sign language recognition

Before RPI, at the Systems and Software Lab at IUT, I worked on Bangla Sign Language. Existing Bangla sign datasets were RGB-only; we built the first complete depth-based dataset of all 47 static signs and showed that depth features from hand skeletons improve recognition.

Data in Brief 2023First author

BdSL47: A complete depth-based Bangla sign alphabet and digit dataset

S M Rayeed, Sidratul Tamzida Tuba, Hasan Mahmud, Mumtahin Habib Ullah Mazumder, Saddam Hossain Mukta, Kamrul Hasan
Data in Brief, vol. 51, 109799 (2023)

The first complete open-access Bangla Sign Language dataset with depth: 47 one-handed static signs (37 letters, 10 digits) from 10 signers, 47,000 images with 21 MediaPipe hand keypoints each. An ANN on the 63 depth-aware features reaches 97.84% F1.

Paper Dataset Code

ICMV 2022First author

Bangla sign digits recognition using depth information

S M Rayeed et al.
Fourteenth International Conference on Machine Vision (ICMV 2021), Proc. SPIE 12084, 2022

A depth-based dataset for the ten Bangla sign digits, built from MediaPipe hand skeletons with x, y and depth per joint. An SVM reaches 98.65%, and the depth features beat the RGB-only Ishara-Bochon baseline.

Paper

DatasetFirst author

BdSL47: A Complete Depth-based Bangla Sign Alphabet and Digit Dataset

Mendeley Data, DOI 10.17632/pbb3w3f92y.3, CC BY 4.0

47,000 RGB images of 47 Bangla signs from 10 signers, with 470 CSV files of 21 MediaPipe hand keypoints (x, y, depth) per image. Signers vary in age, gender, hand shape and skin color, and signs are captured under different scales, rotations, lighting and backgrounds.

Mendeley Data Dryad mirror Code

Other research areas

Fine-grained taxonomic identification Biological visual question answering Knowledge-based VQA and retrieval AI-ready biodiversity data Animal re-identification