Scoping Review Of Sign Language Recognition: An Analysis of MediaPipe Framework and Deep Learning Integration

This scoping review aims to examine the research progression in computer vision about sign language recognition using MediaPipe and deep learning. The following steps are carried out, identifying the research question, Identifying relevant research, selecting studies, organizing the data, and compiling, summarizing, and presenting the findings. The analysis is based on 209 articles published from 2019 to the first quarter of 2024. The findings address two quantitative research questions about the number of titles, cites per year, and number of authors, ultimately concluding that there is a growing momentum in Sign Language Recognition with MediaPipe research, signaling a transition from its initial phase to a more sophisticated and established stage. In a qualitative research question about deep learning techniques, we use filtered methods to analyze the text of articles and search for keywords related to specific deep learning techniques. From 209 articles, we narrowed 38 articles found to analyze, and the result is the most used technique MediaPipe for feature extractor and Long Short Term Memory (LSTM) for temporal features with different configuration layers.