Real Time Air-Writing and Recognition of Tamil Alphabets Using Deep Learning

Writing has always been a prominent way of communication. The way in which the letters are written has been varying with time. From the conventional pen and paper to touch pad and stylus, the way of writing has evolved. Air- Writing is another development in which the characters are written in free space without being limited to a specific tool. This method of writing makes the hand movement easier compared to the conventional methods. Therefore, the air writing and recognition model will be of great help for children who start learning a language. The trajectory of the air written characters is obtained by mapping the focal point using Optical flow in OpenCV. The obtained trajectory is then preprocessed and given to Dense Net 121 which is a type of CNN model widely used for pattern matching along with the dataset from HP labs which contains 3000 images for 11 Tamil vowels. The model which is trained obtained a maximum training and validation accuracy of 98.2% and 91.83% respectively with minimum training and validation loss of 6.35% and 21.04% respectively.