Real Time Implementation of Telugu Character Recognition Using Raspberry Pi

Optical Character Recognition (OCR) technology relies on character recognition, but due to the fact that regional scripts such as Telugu are so complex and intricate in nature, they have not been adequately researched yet. This project proposes a real-time Telugu character recognition system implemented on a Raspberry Pi, aimed at creating a cost-effective and portable solution for regional language OCR applications. The Raspberry 3 model B+ is chosen in this project for its balance between performance and affordability. The dataset used in this project was sourced from IEEE Dataport, ensuring standardized and high-quality data for training. The system accurately identifies individual characters from live camera input utilizing a convolutional neural network (CNN) trained on a dataset of handwritten Telugu characters. The trained model is optimized for avoiding huge overhead so that it becomes feasible to deploy it on Raspberry Pi using TensorFlow Lite for efficient inference. Adoption of advanced techniques such as data augmentation and transfer learning improves the performance of the model in terms of high accuracy and inferencing speed. Benchmarking performance against recent studies indicates an accuracy of 98.2% for Achulu, 94<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> for Hallulu and 98% for Othulu. Through the combination of advanced deep learning with practical application, this project is a step in the direction of OCR solutions for local languages.