Dispositivo assistivo para leitura baseado em captura de imagem e reconhecimento óptico de caracteres

Abstract

This project introduces the development of an assistive device made for people with low vision or dyslexia, utilizing image capture, optical character recognition and text-to-speech in a low-cost portable system. The prototype connects two main modules: The MaixCam, responsible for the image capture and the processing of the OCR, and the ESP32-WROVER, which is in charge of the voice synthesis and the audio transmission to the user via Bluetooth. The communication between modules is made using Universal Asynchronous Receiver-Transmitter (UART) guaranteeing low energy consumption. The Image processing begins with a capture made by the device. To make this image adequate to the OCR it passes through a series of steps of pre-processing. First, the page area is identified analyzing contours and borders, fixing the perspective to align it as a frontal digitalization. After this all external contours and unwanted elements are removed from the background, maintaining only the relevant region of the text. The text is then separated in multiple segments based on the spaces in between the text blocks, generating horizontal cuts that represent different paragraphs. Thus, it is possible to run the OCR in each paragraph separately. Each paragraph is converted into text and sent to the ESP directly, so it can run the Text-To Speech (TTS). The results were that the OCR had two types of accuracy, 92.5% for semantic similarity and 6.25% for exact similarity, which reflects directly on the quality of the TTS. With these results in mind, an additional process of textual post-processing was considered, so that the first accuracy result would become more representative of the TTS’s quality but this process wasn’t possible due to the lack of libraries compatible with the MaixCam. In conclusion, despite the limitations found, this project presents contributions to the field of assistive technologies, introducing an embedded low-cost solution to the problem proposed, reinforcing the potential of automatic reading devices, providing autonomy and inclusion to people with visual impairments and reading disabilities.