Measuring vital signs like pulse rate, breathing rate, pulse rate variability and blood oxidation remotely without using contact electrodes can make continuous physiological assessment possible, both at home and in hospitals. Camera based system to measure photoplethysmographs (PPG) or blood flow waveform (BVP) has been suggested as a possible approach for such cardiovascular assessment. Two major challenge of using camera for recording PPG are - 1) extremely low signal strength of the PPG waveform acquired using a camera, and 2) motion artifact due to movement of the person in front of camera. Thus, all current systems can only extract pulse rate and cannot give reliable measurement of the underlying PPG waveform which could be used for other vital sign measurement and cardiovascular assessment.
In this thesis, I addressed both these challenges First, I devised a novel PPG signal acquisition algorithm which optimally combines PPG signal coming from different regions of the skin to improve the SNR of the system. Second, I track different region of the imaged skin using feature points as it move around during normal activity to compensate for the motion artifacts. Our proposed signal acquisition algorithm gives on an average 4.7dB SNR improvement of PPG signal for people having different skin tones (pale white to brown). Our motion compensation algorithm gives good results for small natural motion scenario like browsing Internet, watching video, and talking on Skype. Thus, PPG waveform comparable to a contact based pulse oximeter can be extracted from the video of a person's face using the proposed algorithm, and have widespread application for remote non-contact cardiovascular assessment.