Automatic Assessment of Infant Sleep Safety Using Semantic Segmentation

In this paper, an infant sleep prevention solution based on semantic to access infant environmental hazards is presented. To promote safe sleep evaluation and implement sustainability in rural underserved communities, we use deep learning techniques to automatically assess photographs of the infant's sleep environment and report unsafe environments. To achieve this, we first built and labeled a dataset of 626 images from infants in various sleep positions and environments. The segmentation architecture is composed of a downsampling path responsible for extracting coarse semantic features, followed by an upsampling path trained to recover the input image resolution and finally, a pixel-wise classification layer. The trained model is also integrated into an android application to provides a sustainable evaluation/assessment tool. We achieve state-of-the-art results and demonstrated that the automated assessment system could identify safe/unsafe sleep environment using photographs.