Smart Glasses to Detect Obstacles and Provide Assistance for Visually Impaired People

Our aim is to introduce a new design of smart glasses that can provide assistance in multiple tasks while maintaining at a low building cost. Among all assistive devices, wearable devices are found to be the most useful because they are hand free or require minimum use of hands. The most popular type is head mounted device. Their main advantage is that the device points naturally at the viewing direction, thus eliminates the need of additional direction instructions, unlike other devices. This product focuses on people who have hearing abilities. The general principle of operation for such glasses is by giving instructions via switches and listening to the output through an earpiece. The design depends mainly on the processing unit, which is the raspberry pi 2, in this case. The main hardware is a Linux based ARM processor that accepts a micro SD card and thus allows us to increase the number of task functions as we wish. A raspberry pi camera was used for image acquisition. It was connected to the raspberry pi using a ?ex cable, and was ?xed on the top middle of the glasses for optimal image capturing. The raspberry pi has an audio port which connects to earpiece. The raspberry pi GPIO port was con?gured to receive input from push button switches. To identify the text easier, the reading material is placed within a customly-designed frame with red borders. The general principle of operation for such glasses is by giving instructions via switches and listening to the output through an earpiece. Similarly in this case, the user starts the task mode by a push of the button. For text recognition mode, the glasses will ?rst con?rm if the text area is correctly positioned and readable. Otherwise, it will ask the user to change the orientation of the material. After con?rmation, the view is processed in real time to get the image sent to optical character recognition (OCR) software for text extraction and subsequently forwarded to a text-to-speech synthesizer. The text is then read through the audio output port. In the reading mode, the main challenge is the image quality, text position and orientation in the image. Therefore, the ?rst step is to detect the red borders and the frame orientation. To simplify subsequent image processing, we propose an indicator to inform user if the image is skewed signi?cantly or part of the frame is cropped. Once the text area is localized and cropped, image is enhanced by noise ?ltering, contrast enhancement (histogram matching technique) and morphological operations.