Improving Motion Robustness of Contact-less Monitoring of Heart Rate Using Video Analysis

Heart rate is the number of heartbeats per unit of time. Heart rate is one of the important physiological signals used by medical professionals to assist in the diagnosis and tracking of patients medical conditions. With every heartbeat a blood pressure pulse travels from the heart to the other body parts. Typically, it is referred as Blood Volume Pressure (BVP) or cardiac pulse or only pulse. A heart rate of a person can be detected if a cardiac pulse rate is detected at any other body part. Conventionally, a medical person measures cardiac pulse at any place that allows an artery to be compressed against a bone, such as at the neck, at the wrist, behind the knee, on the inside of the elbow, and near the ankle joint. Conventional automatic measurement of cardiac pulse include Electrocardiography (ECG), finger clip or earlobe pulse oxymetry sensors and chest strap. All above specified methods needs a direct contact with patients body to measure heart rate. Remote and non-contact measurements of the cardiac pulse can provide comfortable physiological assessment without clutter of wiring and electrodes. Photo-plethysmography(PPG) is a simple, non-invasive and low-cost optical technique that can be used to detect blood volume changes in the microvascular bed of tissue. Photo-plethysmographic signals are measured remotely (> 1m) using ambient light and a simple consumer level digital camera in a movie mode. Heart rate can be quantified up to several harmonics using PPG method. However, extracting the physiological signal from a video is affected by the motion of a subject. In literature, the algorithm developed by Division of health science and technology, Harvard-MIT claims to be the first motion improved algorithm to detect heart rate from a video analysis. In our study, we investigate above mentioned algorithm in detail and quantify its performance. The algorithm is verified for the motion robustness it claims using videos of a subject with different types of motion. It was found that the algorithm performs satisfactorily under stationary, slow translation and rotational motion with small rotation angle. Improvements in the existing motion tracking scheme used by the algorithm and signal processing of the algorithm are suggested, implemented and quantified. The improved motion tracking scheme involves skin pixels detection and patch tracking in successive frames of a video. To exploit further improvements, changes in a way signals are extracted from a video are suggested and verified for the performance improvement. It was found that a typical combination of red, green and blue traces gives better signal to noise ratio than the other. The reason why the particular combination works better than the others is also investigated from the light absorptivity curve of Oxyheamoglobin and De-oxyheamoglobin. Further improvement in the performance of the algorithm is achieved by extracting the differentiated signals from a video. After series of evolution in the existing algorithm, the final algorithm is more robust and computationally less expensive than the existing algorithm.