Automated Virtual Character Detection System from Live Video Streaming using Deep CNN

Automatic character recognition (ACR) transforms scanned document images into a language that computers can understand and recognize text that appears in photos. The most recent technological developments have highlighted the necessity for reliable automatic character recognition techniques. There was a need for a more contemporary recognition architecture. Thus, character recognition is accomplished using deep learning methods in this study. Finding the ideal set of features and then giving them to the classifier helps solve a variety of artificial intelligence tasks. Finding the ideal combination of characteristics can be challenging for many activities. The solution to this problem is deep learning which outperformed basic AI models. Proposed work has tried multiple experiments to find best performing model to recognize Devanagari Numerals through virtual representation. In this work, normal and complex CNN both types of models have been used to train and test purposes. The suggested model has produced 97.8% accuracy and used at the backend to recognize Hindi letters virtually indicated by the user in front of their webcam, also the model predicted ‘’ with 92.6% and 96.30% respectively at different instances during experiments. This work can be utilized as fun game for children to learn any language easily.