<![CDATA[A blind person generally has limitations in visual sensing and uses aids such as a cane to assist their mobility but still has obstacles, especially when they are in a building and want to find the room they are aiming for. Room labels that are usually printed and cannot be recognized by a blind person can be identified with the help of a portable device that performs computing in the field of computer vision. By relying on the MobileNet SSD algorithm, which can detect the presence of room labels with a fast computation, and Optical Character Recognition (OCR) which can recognize the name of the room label, users can hear the name of the room spoken through the speakers. In short, the system converts visual information into audio information that a blind person can receive. Even though the primary processor is an edge device such as the Raspberry Pi 3B+, an additional Intel Neural Compute Stick 2 accelerator device can help the detection process go faster because the detection algorithm involves a computationally intensive deep neural network. Based on the tests carried out in this study, the room label detection test using MobileNet SSD resulted in an accuracy rate of 80% with an average computation time of 68.44 ms. While for recognition using OCR, it produced an accuracy value of 93.65% with an average computation time of 263.05 ms. In addition, the integration results based on digital image input with sound output obtained an accuracy rate of 50% because the sound is only pronounced if the recognition results match the name of the existing room label.]]>
Copyrights © 2022
