Heart beat rate calculation from facial video recording on smartphones

Abstract

As each generation of hand-held devices is released, their hardware continuously improves at a near-exponential rate. Due to their widespread popularity, people have become increasingly reliant on their personal hand-held devices and routinely expect the next generation of devices to be able to support a wider range of applications. One such application could be an application that extracts a person's heart rate through a real-time video feed of a person's facial region without requiring any additional equipment. Presently, the main methods for accurate heart rate measurement require some form of specialised equipment that often tend to be too costly for the general public and therefore end up getting disregarded. This means that an accurate, free and user-friendly mobile- device application could easily become one of the main methods for personal heart-rate measurement. In this Final Year Project (FYP), an Android application that can detect the subtle changes in skin colour from frame to frame of a live video recording of an individuals face caused by the blood circulation in the capillaries close to the skin was developed. Human vision does not have the spatio-temporal capabilities needed to detect these minor changes, however as blood ows into these aforementioned capillaries the skin becomes redder and vice-versa as blood flows out. By amplifying these near invisible colour changes over time, the number of times blood ows in and out of the facial region can be calculated. These results can then be used to nd the Beats Per Minute (BPM). A recently proposed algorithm, known as Eulerian Video Magni cation (EVM) is an algo- rithm capable of detecting these minute changes in colour. The algorithm designed in the original paper works only in an environment with a stable background and with constant lighting whilst also requiring a few minutes of processing time before an amplifi ed video is shown [1]. The main aim of this FYP is to create a toned down version of the original EVM al- gorithm that is capable of amplifying the minor colour changes in a facial region caused by blood circulation in a real-time contact-free manner. Unlike the original paper, the proposed solution is expected to obtain correct results in any realistic condition, and not just in a room with a still background and constant lighting conditions. After evaluating the results, the developed application reported heart rates nearly identical to a finger monitor with an average accuracy of 97% over all tests.