Digital Empowerment For The Visually Impaired

Abstract

This project aims at developing a portable device assisting the visually impaired people to recognize bus number in real time. The limitations for the traditional holding board method are that a blind person cannot acknowledge the presence of in-coming buses and they may hold a board for a long time. These limitations could be solved by my proposed equipment using computer vision technology and hardware integration. My proposed device was composed with several modules including a Raspberry Pi 3 A+, a web camera, a Bluetooth headset and a power band. With the 5-steps workflow as a backbone, the system was successfully developed. By using machine-learning-based object detection algorithm – Haar Cascade Classifier, it was able to detect a bus front object with 100% precision and broadcast the bus number with 60% precision. With a series of image processing strategies, the text region could be cropped with specific location. Transforming the information from image format to text format, a trained Tesseract OCR could perform text recognition with 77% precision. Lastly, the bus number information would be delivered to the user by mean of eSpeak text-to-speech function through a Bluetooth headset module. This device could perform the following functions: 1. Recognizing route label of in-coming double-decked buses in Hong Kong; 2. Being notified with a Bluetooth headset; 3. Operating in a hands-free manner, i.e. no holding is needed