Non-contact breath cycle analysis for different breathing patterns using RGB-D videos

Abstract

Traditionally, yogic breathing exercises have been considered an effective adjunct in managing stress, chronic pain and depression. Real-time bio-feedback based gaming applications can provide an opportunity to broaden the reach of yogic breathing exercises. To extract the bio-feedback parameters on human breathing behavior, robust real-time quantification of inhalation, exhalation, and retention durations for each breath cycle is very crucial. This paper implements the extraction of every respiration cycle of a person along with the duration of inhalation and exhalation using RGB-D videos under minimal movement. The proposed method uses the abdominal-thoracic region as the region of interest (ROI) and relates the small displacement of the chest region with breath and obtains a time-series signal that represents the user's breathing pattern. Appropriate digital filtering techniques are employed in eliminating the noise components and a novel algorithm is proposed to extract individual breath cycle parameters. From the extracted parameters, a gamified bio-feedback application to regulate respiration in real-time is demonstrated. The proposed methodology is evaluated for different breathing patterns against a collected representative dataset, which includes ten participants of different age groups and gender. Results indicate that the breathing rate can be detected with a mean RMSE error of 0.435 s and 0.467 s for the inhalation and exhalation times between the proposed method and the ground truth measurement.