Page 767 - AI for Good Innovate for Impact
P. 767

AI for Good Innovate for Impact



                   Use Case 8: AI-Powered Indian Sign Language (ISL) Detection and

               Translation                                                                                          4.9: Accessibility


















               Organization: International Institute of Information Technology, Hyderabad.

               Country: India

               Contact Person:

                    Pandey, Aishani, aishani.pandey@ research .iiit .ac .in
                    Sachdeva, Arush, arush.sachdeva@ research .iiit .ac .in
                    Mathur, Vivek -vivekofficialwork1@ gmail .com


               1      Use Case Summary Table

                Item              Details

                Category          Accessibility
                Problem           There is a big communication gap between people who speak ISL and
                Addressed         those who don’t. For the deaf community, social integration and informa-
                                  tion access are hampered by the absence of real-time translation tools.

                Key Aspects of  Data Acquisition: Gather and preprocess Indian Sign Language(ISL) data-
                Solution          sets from a variety of pertinent and helpful open-source sources, such as
                                  expert-curated databases, publicly accessible sign language resources,
                                  and YouTube tutorials. To ensure accurate representation, specific context,
                                  vocabulary, phrases, and videography protocols related to the ISL context
                                  will be applied during data curation.
                                  Tech Stack: Python (Pandas, Open Computer Vision (OpenCV)), YouTube
                                  Application Programming Interface (API), Web Scraping (BeautifulSoup),
                                  Video Annotation Tools (e.g., Visual Geometry Group (VGG) Image Anno-
                                  tator)
                                  Data Enhancement: Augment data for low-resource variations using pose
                                  estimation, synthetic data generation, and adversarial augmentation tech-
                                  niques to improve generalization.
                                  Tech Stack: TensorFlow, PyTorch, OpenPose, Generative Adversarial
                                  Networks (GANs) for augmentation













                                                                                                    731
   762   763   764   765   766   767   768   769   770   771   772