| Summary: | 碩士 === 元智大學 === 電機工程學系 === 104 === Non-contact equipment for measuring physiological signal is a challenge task. Especially, to track multi-targets respiratory rate is much more difficult due to the vertical motion from upper-body is not conspicuous. Respiratory rate is an important physiological signal which is highly correlated to vital signs, diseases, and emotions. Therefore, a real-time robust image-based technique is developed for measuring multi-targets respiratory rate variation. In this work, we propose an ultrafast method to detect human body using depth information only. The front/back body of humans can be detected. In addition, under the demand of computation efficiency the salient area is found by Haar-like features from upper body. The vertical body motions estimated from optical flow through consecutive frames are decomposed by a median motion filter and smoothed by local mean filter. Finally, the respiratory rate is calculated by zero-crossing method. For performance evaluation, four kinds of experiment are conducted. The experiment results show that our proposed system for measuring respiratory rate has achieved a high accuracy using an extensive dataset.
|
|---|
