Performing bidirectional LSTM-based machine learning
Extracting feature points from the hands, face, and mouth of sign language implementers
On the 11th, the Korea Electronics Technology Institute (KETI) announced that it had developed Korea's first artificial intelligence-based sign language recognition technology through the government's 'Autonomous Intelligence Digital Companion' project.

The developed technology performs machine learning based on bidirectional LSTM (Long short-term memory), a type of recurrent neural network. This is a technology that extracts features from the hands, face, and mouth of a sign language implementer and interprets them into Korean sentences. Compared to other technologies, it has high practicality, simplicity, and accuracy.

To develop this technology, KETI has been working with Professor Byeong-Cheon Yoon's team at Nazareth University, Korea's leading institution for sign language interpretation research, to build sign language data for artificial intelligence learning since 2017.

“We hope that through the development of this technology, artificial intelligence will serve as a bridge connecting the world so that no one can experience inconvenience,” said Jeong Hye-dong, director of the KETI Artificial Intelligence Research Center, who oversaw the research. “In the future, we plan to expand the application of AI-based behavioral recognition technology to intelligent CCTVs for use in urban safety, or to equip self-driving cars so that they can even recognize police officer hand signals, thereby upgrading them to highly intelligent AI technology that understands all ‘words and actions.’”