Inspiration
Our inspiration was two-fold. Firstly, we realized that traditional baby monitors are excellent at alerting parents of their babies crying, but they are entirely unable to alert parents of their child stopping breathing. While this may generally be assumed to not be a problem, Sudden Infant Death Syndrome (SIDS) is a pathology which claims the lives of over 1,700 infants every year in Canada alone.
With this problem in mind, we were inspired by Wu et. al's 2012 paper titled Eulerian Video Magnification for Revealing Subtle Changes in the World. In it, a novel video processing technique was used to amplify subtle changes in subject, including small movements caused by respiration.
What it does
The current version of SlumberSafe is able to track breathing in subjects through real-time video monitoring and recorded video analysis. This information is stored and can be analyzed by parents and health professionals to non-invasively assess the health of the baby and the quality of their sleep.
How we built it
We the YOLOv3 subject detection model to track the precise location of the subject. Using this, we detected changes in pixels intensities in the subject between frames and applied thresholding to detect whether the subject was moving or immobile. By averaging the prediction across clusters of frames, we could detect inhalations and exhalations in adults, and exhalations in infants.
Challenges we ran into
Signal processing is very difficult. We were not able to implement the Eulerian Video Magnification in the paper which involved both spatial and temporal processing. This was because further signal processing was necessary to isolate the vitals signals from the noise.
Accomplishments that we're proud of
We're proud to have a fairly accurate breathing rate despite the challenges of processing the video signal.
What we learned
Video is very noisy and extracting small signals can be exceeding difficult. Moreover, filtering can be used effectively to track movements at specific frequencies, but overly aggressive filtering will result in tracking noise rather than true signal.
What's next for SlumberSafe
With more time, and proper sleep, we hope to fully implement real-time Eulerian Video Magnification to obtain more reliable signals for breathing, and in the future, even for pulse. Moreover, we hope to implement a signal processing pipeline to dynamically adjust the movement thresholding values to properly track vitals in different environments.
Log in or sign up for Devpost to join the conversation.