A real-time ASL alphabet classifier capable of transcribing ASL alphabet signs to text.
In a Data Mining course taken during my undergraduate studies at the University of Georgia, we were given free rein to conduct mini research projects using some of the machine learning techniques studied in class. Fresh off of a study of convolutional neural networks and their use in image classification tasks, I devoted one mini project and my final term project to the study of automated ASL alphabet sign classification using machine learning techniques.
These “mini” research projects were indeed miniscule, conducted over the course of only one to two weeks for the mini project and three to four weeks for the final term project. Accordingly, the scope and impact of these research projects are considerably limited and indeed many shortcomings exist in these sister works. Most importantly, and quite strikingly, the projects suffer a great burden of weak training and testing data which, to no surprise, severely limit the ability of the resulting classifiers. With more time, one can hope that the ability of the developed classifiers would be greatly improved by the collection and careful processing of much more and much higher quality training and testing data. Nonetheless, they were a fun exercise into data mining and machine learning and have produced a silly, but fun ASL transcribing tool which I have affectionately named ASLScribe.
With these disclaimers out of the way, I present two sister works on ASL alphabet classification. The first is a (very) mini project investigating the use of convolutional neural networks on the task of ASL alphabet classification. The second is a slightly more involved term project, building off of and extending the findings of this first mini project. Each project has an accompanying research paper which can be accessed using the links below.
Mini Project: ASL ClassificationFinal Term Project: ASL Transcription
The term project concerns itself with the production of an ASL transcriber using the best classifier found in the paper. This transcriber, the aforementioned ASLScribe, is hosted on this website and can be accessed below.
