A prototype sign language translation web application made for my baccalaureate's research project.
This app translates basic Spanish Sign Language fingerspelling signs to text using Mediapipe HandPose and an implementation of a Naive Bayes Classifier.
A 50-page document written in Catalan explaining all the inner workings and showcasing the final product can be found here.
app.mp4
These are some of the possible improvements and features that could be implemented if the project was updated.
Instead of passing the hand's points directly to the classifier, which makes classification errors when the hand is rotated, image features could be extracted using the 7 moments of Hu.
Instead of storing the data as plain text files with numbers, data could be stored as raw binary files containing floating point numbers.