Video based actigraphy and breathing monitoring from the bedside table of shared beds

Good sleep is an important factor for a high quality of life. Presence of sleep disorders requires patients to undergo polysomnography examinations at a sleep clinic, which involves attaching multiple head and body sensors. This results in an uncomfortable and unnatural sleep setting for the patients, complicating an accurate diagnosis. Currently, this diagnosis is done manually, making the entire process time consuming and cumbersome. We propose a camera based system capable of analyzing the subjects in their own sleeping environment. The system segments the primary subject from the background and any other bed occupants, then computes the actigraphy and the breathing characteristics of the subject. Segmentation is performed with an AdaBoost classifier using among others motion, intensity, focus and Histogram of Oriented Gradients features. A sum of absolute differences operation on the pixels within the segmented area of the primary subject returns the actigraphy signal. The breathing characteristics are extracted based on small motion analysis of consecutive and reference frames. The proposed system has been evaluated on four healthy adults for actigraphy and on five healthy adults for breathing analysis using a Texas Instrument Chronos wrist watch and inductive respiratory belts as references respectively. Evaluation was performed using the metrics accuracy, precision, sensitivity and breathing rate correspondence. The proposed system has an average accuracy of 88 % at a precision of 79 % for segmentation of the primary subject. It can detect movements up to an accuracy of 85 % while outperforming wrist actigraphy at 75 % accuracy. State-of-the-art video based breathing algorithms were surpassed with an overall sensitivity of 87 %, precision of 90 %, and a breathing rate correspondence of 93 %.