Abstract

Video data is now commonly used for analysis in surveillance, security, medical and many other fields. The development of low cost but high-quality portable cameras has contributed significantly to this trend. One such trend includes non-invasive vital statistics monitoring of infants in Neonatal Intensive Care Units (NICU). National Center for Health Statistics Publications has reported a high infant death rate (23,215 in 2014). This statistic has drawn the interest of health system professionals. Due to occurrence of conditions like bradycardia, apnea and hypoxia, these preterm infants are kept in an NICU for constant monitoring. One of the problems faced at the NICU is the use of traditional sensors for vital statistics monitoring which might cause damage to the already fragile skin of these infants. A contact-less approach to record such vital signs can now be employed using a live video feed. Recent literature shows that it is possible to extract features from videos that are invisible to the human eye, employing various image and signal processing techniques. One of the algorithms demonstrated recently is the extraction of medical vital signs based on wavelet filtering of monochrome video data.

The pumping of blood to various parts of body from the heart in rhythmic fashion causes subtle changes in the skin tone of humans. These changes are periodic in nature as the pumping action itself is periodic corresponding to the heart beat. Typical heartbeat of a human ranges from 50-200 beats per minutes (bpm) implying the range to be 0.83-3.33 Hz. Similarly, a video is a sequence of frames with frame rate ranging from 15-30 frames per seconds (fps). This creates the possibility of using videos to detect heart rates as the Nyquist Criterion is met with ease. The subtle changes in skin tones can be further processed and magnified. The mean gray level signal obtained from such a process has been found to be resembling the pulse rate waveform obtained from photoplethysmograph(PPG) sensor to measure pulse rate. The other approach is to use color channel domain like HSI (Hue, Saturation and Intensity).

With the above concept, a video processing algorithm was designed in MATLAB. Short videos of several subjects with different skin tone were recorded for the analysis. In order to compare to the ground truth, pulse data were recorded at the same time using the photoplethysmographsensor of a wearable watch. Upon implementing the algorithm designed, on the videos, it was possible to extract waveforms from the videos that resembled the pulse waveform recorded from the ground truth measuring device. The percentage error was in the range of 0.2 to 1.4%. This led to the conclusion that video data can be analyzed to extract heart rate and with further study can be used for real time monitoring of cardiac activity of infants at an NICU.

Date of publication

Fall 12-10-2018

Document Type

Thesis

Language

english

Persistent identifier

http://hdl.handle.net/10950/1212

Committee members

Dr. Mukul V. Shirvaikar, Dr. Premananda Indic, Dr. Ron J. Pieper

Degree

Master of Science in Electrical Engineering

Share

COinS