| Formatted contents note |
ABSTRACT: STATEMENT OF THE PROBLEM: The BIRD (Braille Image-Based Recognition Device) project aims to develop a device that can accurately identify braille characters through visual recognition and provide an audio output. The study has specific objectives, including a system for accurate braille character recognition using machine learning algorithms, creating training models for identifying braille characters, and testing and evaluating the functionality of the device. The study seeks to answer questions related to the device’s accuracy, the effectiveness of the object detection algorithm under different conditions, and the audio output’s alignment with the detected braille characters. The general objective of the project is to address the braille literacy crisis and promote braille familiarity among visually impaired individuals. RESEARCH METHODOLOGY: The research design is based on a quantitative approach, involving the collection and analysis of numerical data. The study aims to evaluate the performance of a developed braille recognition device through tests conducted in different situations. Descriptive statistics will be used to analyze the data and provide insights into the device’s performance. The study also includes a user testing survey questionnaire that focuses on the ISO/IEC 25010 quality model. The questionnaire assesses the functional suitability, performance efficiency, and usability of the device. It evaluates factors such as functional completeness, correctness, appropriateness, time behavior, resource utilization, capacity, recognizability, learnability, operability, user error protection, user interface aesthetics, and accessibility. The research locale involves conducting a preliminary interview with a chosen professional in the field of Braille, along with user testing of the device at Pamantasan ng Lungsod ng Maynila in Metro Manila, Philippines. The device is optimized for reading braille alphabet characters and is primarily targeted at braille beginners. The software components include managing datasets, training the object detection algorithm, deploying the model, and developing the web application. The hardware components consist of the ESP32 camera and ESP32-CAM-MB as the main input sources. The software integration involves platforms like Roboflow and Google Colab, while the hardware integration connects the camera module to the ESP32-CAM-MB board. The working process of the device involves the user reading a braille document, which is recognized and translated into natural language by the device. The user has control over the device’s output, and the device can read the braille text again upon user activation. The image processing process includes gathering braille character datasets, classifying them, training a model, and integrating it with roboflow.js and hardware components to enable image processing. SUMMARY OF FINDINGS: The findings indicate that the object detection model used in the device achieved high accuracy, improved bounding box predictions, and proficient object detection and classification capabilities. However, the training dataset was found to be imbalanced, which may affect the model’s performance on specific object classes. The device’s accuracy in recognizing braille characters varied depending on the type of braille, which embossed braille flashcards showing the highest accuracy. The performance of the device was influenced by factors such as lighting conditions, size and color of braille characters, and training data. The device had a longer detection range for embossed flashcards compared to other types of braille. User testing revealed positive feedback on the device’s functionality, performance, and usability, indicating its potential as a valuable tool for learning braille. However, there were suggestions for improvements, including word output functionality, enhanced number detection, and improvements in camera functionality and design. CONCLUSION: The Braille Image-Based Recognition Device (BIRD) is designed to translate braille into natural language. It produces highly accurate translation, with median accuracies of 73% for embossed braille documents, 88% for embossed braille flashcards and 80% for digital braille. However, certain braille characters may be more challenging to detect due to limited training data. The device’s object detection algorithm performs differently in various lighting conditions, with better detection of embossed braille flashcards under low-key lighting and the best performance in detecting embossed braille flashcards under high-key lighting. The device has a long maximum range for detection, which is beneficial for its use as a headband attachment. User testing shows positive feedback on the device’s ability to produce audible representations of braille alphabet characters. The BIRD device shows promising results as a supplementary tool to aid in addressing the braille literacy crisis and promote braille usage without replacing teachers. It serves as an optical braille recognition assistant tool that supports visually impaired individuals and beginners in their journey towards braille literacy. RECOMMENDATION: The researchers have made several recommendations for future development of the Braille reader device based on feedback from users. They suggest improving the audio output, using a better camera, enabling the model to form words from predictions, upgrading the device’s aesthetic, using dedicated batteries for portability, improving dataset quality and quantity, exploring different object detection models, increasing the number of users to test the device, and creating a standalone device. These improvements are aimed at making the device more accurate, usable, and user-friendly for visually impaired leaners. |