Deep Learning on Smart Glasses

This project was done as part of one of my subjects in my masters degree. It was given by the Center of Integrated Emergency Management (CIEM). CIEM is a multidisciplinary research center which conducts research in fields like networks, mobile devices, human-centered sensing, social media, decision support and much more. It is a top research priority at the University of Agder.

Project Goal

The goal of this project is to create a proof-of-concept application for the BT-200 smart glasses that demonstrates how deep learning algorithms for computer vision can be applied to smart glasses. The app should take pictures with the camera mounted on the glasses, classify these images, and display the classification results on the smart glass display in real-time. We will also investigate how various physical surroundings and lighting, may affect the classification performance with input from the smart glasses. This can give an impression on how a dataset from the smart glasses performs on the current state-of-the-art deep learning algorithms. Also, extensive testing identifies problems early in the project, which is useful for future work.


This project was quite successful. We created an app that runs independently on the glasses. It takes one picture every two seconds and sends it to a server. This server which uses the framework Caffe, classifies the image and returns the top three results that are displayed on the glasses. CIEM has show a grate interest in our prototype. They are now using our findings for further research and building a more powerful and stable version of our prototype.

The full report: Deep_Learning_on_Smart_Glasses