Using neural networks to enhance the quality of ROIs for video based remote heart rate measurement from human faces

Classical approaches for measurement of a patients heart rate (HR) inherent several disadvantages like discomfort or irritation of the skin. Therefore, non-invasive and non-obtrusive methods, like video based ones become more and more popular. A majority of the methods described in the past, use video data of human faces. These methods recon minimal changes, invisible for the human eye, in the color spectrum of a persons face to measure the heart activity. It is obvious, that measuring the HR from a persons video data is not a trivial – though a very challenging – task, as seen in the last years. Quality of the results can be improved in multiple ways. Most of the presented approaches using filter methods like independent component analysis (ICA), Blind source separation (BSS) or many others to improve the given data. Because all of these techniques intervene at a relative late point in the algorithm, in this paper another approach is described. By detecting and enhancing the region of interest (ROI) from video data made before, it is possible to improve quality of the given data for the later use in any algorithm. A huge acceleration in processing time is realized. To achieve this the here proposed method using neural networks to detect and improve the ROI from given video data is used. By the later use of ICA the here proposed algorithm is able to measure HR with a very high accuracy. The combination of this two techniques, is able to deal with various situations in different lighting conditions and patients activation level, therefore a better accuracy and an improvement in runtime towards more realistic applications was realized.