Abstract
Heart rate (HR) is one of important indicator for human physiological diagnosis, and camera can be used to detect it via photoplethysmograph (PPG) signal extraction. In doing so, number of sample images required to measure the HR signal, and quality of the images itself are important to yield an accurate reading. This paper tackles such an issue by analyzing the effect of sampling interval to HR reading in compressed and original video format, obtained in various ranging locations. Technically, important facial points from video stream were estimated by using cascade regression facial tracker. Based on the facial points, region of interest (ROI) was constructed where non-rigid movement is minimal. Next, PPG signal was extracted by calculating the average value of green pixel intensity from the ROI. Following that, illumination variation was separated from the signal via independent component analysis (ICA). The PPG signal was further processed using series of signal filtering techniques to exclude frequencies beyond range of interest prior estimate the HR. From the experiment it can be observed that sampling time of 2 seconds in uncompressed video shows promising HR within the range of 1 to 5 meters.
Highlights
Over the past years, many methods introduced to monitor human Heart rate (HR) reading since it closely relat ed to human physiological aspects
We showed the effect of uncompressed video to the HR reading
The PPG signals were extracted from eight different subjects that was recorded at 60 FPS from three different distances of 1 meter, 3 met ers and 5 meters and producing 24 video samples
Summary
Many methods introduced to monitor human HR reading since it closely relat ed to human physiological aspects. ISSN: 2302-9285 provide more accurate outcome for PPG signal extraction because haemoglobin light absorption is most sensitive to oxy-generation changes for green light com pared to blue and red lights [5, 9] Another interesting works that had been introduced is pulse detection via head motion [10] that showed promising outcome for translating subtle head motion into HR estimator. Work by Zha o et a ll a lso suggest tha t there a re deteriora tion in PPG signa l amplitude, SNR and signal trace due to video compression [18] Their findings about video duration is very interesting, motivated us to analyse more by integrating the minimum time sampling requirements to the compressed video for measureing the non-contact HR reading
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call