Design and Implementation of AI Blind Glasses based on Raspberry Pi - Yang Zhijie

Hi everyone！I’m Zhijie Yang, from the RDFZ Chaoyang Branch School. This summer, I worked as a volunteer at Bethel, a national federation of blind children. Seeing all the inconveniences of blind life, I came up with the idea of making a pair of Smart Glasses that could help blind children to live better.
This Smart Glasses consists of Raspberry Pi, camera, Speech Recognition module, expansion board and batteries, and these components are integrated into a helmet. The Smart Glasses can identify items within 1-meter distance, you just ask "What is this," and it will give you an answer.
Now with the Smart Glasses, when they want to find something, they just need to walk by the table or cabinet, through the sound prompt, they can find the item.
This is how the Smart Glasses works. On the Raspberry Pi I have developed and deployed a deep learning-based object detection and recognition system, with PyTorch, OpenCV and CNN. I selected the most commonly used items in blind children’s daily life, like water cups, bowls, spoons, tissues, toys, and built a dataset containing 5,000 pictures, and labeled these data to form a detection dataset. I generated classification models with some public dataset like ImageNet, and thus determined the basic parameters of the detection model.
However, this is the version 1.0 Smart Glasses, which have many defects, such as the detection speed is not fast enough, the detection accuracy of some items is not high enough, the types of items is limited, and the Raspberry Pi often go overheating.
In the future version 2.0, I hope to resolve the defects. Let me know if you have any good ideas, will appreciate it, thanks.

[@user-mf6ug4te2d]

good