Real-Time Detection of Bangla Sign Language for Shopkeeper-Customer Communication

This data set provides a sample of 6480 labeled images of still hand gestures depicting 14 distinct signs of the Bangla Sign Language (BdSL). It is specifically geared towards helping to advance the real-time communication tools to the deaf and the hard-of-hearing people and specifically in the business world like shopkeeper-customer relationship. The data was formed under the free will of people of different ages and types. The images were shot through the use of different smartphone-specific camera (13 MP and above), in varied real-life conditions. The light conditions include changes in light environment including natural light and dark light, the position and the backgrounds. All the cameras operated the auto exposure mode and had standard settings without manual settings and filters.

Each image in the dataset corresponds to one of 14 commonly used Bangla words or phrases:
"আমি" (I), "আপনি" (You), "স্যার" (Sir), "প্যাকেট" (Packet), "বিস্কুট" (Biscuit), "খাওয়া" (Eat), "এক" (One), "দুই" (Two), "তিন" (Three), "চার" (Four), "পাঁচ" (Five), "ওজন" (Weight), "টাকা" (Money), and "আমি তোমাকে ভালোবাসি" (I love you).
The data is split into two groups in order to fulfil the requirements of different machine learning reasons:
• Training (Detection): This folder consists of annotated images (bounding box) that is used to train object detection models such as YOLOv10.
• Testing (Recognition): It has images that are not labeled and images that are labeled with the classes and may be used to test and train a gesture classification model.
They are all in JPG format and have been filtered and compressed to make the file size smaller and still maintain quality of the image. This brings the dataset to be more available to the researchers with limited resources of computation.
The current data is especially useful to those of us in the sphere of:
• recognition of sign language
• Human to computer interaction
• Supporting technologies of the deaf and hard-of-hearing population
The fact that the dataset allows to accurately identify and detect gestures of the Bangla Sign Language enables inclusiveness to its members and closes the communication gap between the hearing impaired and the rest of the society. It can also be used as a benchmark corpus on future study on regional sign language systems, which are usually underexamined in the global datasets.