Camera-Based Method for Respiratory Rhythm Extraction From a Lateral Perspective

This work proposes a new method based on computer vision algorithms to measure the respiratory rhythm of a subject from a lateral perspective. The proposed algorithm consists on tracking the motion of the intercostal and abdominal muscles by the means of dense optical flow, being the novelty of the proposed method the extraction of the respiratory signal from the phase of the optical flow, while extracting at the same time a quality index from the modulus. 15 healthy subjects were measured while seating, and 4 tests were performed for each subject involving different scenarios. The algorithm has been validated using a commercial wearable thorax inductive plethysmograph system. The instantaneous frequency for the constant frequency respiratory tests, and the breath to breath analysis and instantaneous frequency of the free breathing test have been computed to assess the performance and error of the proposed method for the respiratory acquisition. Finally, a statistical analysis has been performed to assess the accuracy and performance of the quality index. The results of the study show a high agreement between methods in the 0.1 Hz and 0.3 Hz test. For the Free breathing test, both the cycle by cycle and Instantaneous frequency results show a low error between methods with high sensitivity in the cycle detection. The hypothesis that the modulus of the optical flow could be used as a quality index has been corroborated, with very good statistical results. Moreover, due to the simplicity of the proposed algorithm, the proposed method can perform in real-time while measuring respiratory rhythm and assessing the quality of the acquired signal. Further studies taking into account external vibrations have to be performed, to assert that the proposed method can be used in demanding conditions.