Neural Networks for Improving Wearable Device Security

Wearable devices with first person cameras, such as Google Glass, are becoming ubiquitous. Additionally, the importance of security will increase as these devices are likely to capture private or sensitive photos. In this paper we introduce a Convolutional Neural Network (CNN) which is capable of detecting various security risks with relatively high accuracy. Our approach is also capable of classifying images within approximately half a second–fast enough for a rapid, mobile environment. We also investigate numerous routes for improving accuracy with our CNN classifier including object segmentation and artificial data generation. Finally we verify our approach using a manually collected dataset which was tested using two different Android devices.