The Study Of Non-contact Detection Method Of Heart-rate Based On Canonical Correlation Analysis

Heart rate is an important indicator of cardiovascular health of the human body.Commonly, heart rate can be measured directly from pulse, electrocardiogram(ECG)and blood volume. And some non-contact ways also have been proposed, such as the technology of non-contact electrodes, magnetic induction, microwave radar which provide an effective tool for non-contact measurement. However, all these methods have the shortcomings of expensive equipment, inconvenience operation, causing discomfort easily and so on. In addition, it cannot satisfy the requirement of growing daily home monitor and intelligent medical. Non-contact heart rate detection method based on video has many advantages compared to the others. For this video-based method, its main algorithm is independent component analysis(ICA) followed by periodic analysis in frequency domain which also has some shortcomings like compared low accuracy, long cost of measurement time and some restrictions when subjects are in motion state. Considering these issues, in this study we anticipate to propose an effective and reliable detection method of heart rate which can be also worked in the motion mode. Taking into account the convenience of video image acquisition, multi-subject measurement simultaneously was also concerned. These aspects have been studied as following:First, many factors during the process of extracting heart rate from the video images had been considered. Some pretreatments of images have to be involved in order to obtain a reliable result of heart rate. The problem of uneven ambient lighting was solved by homomorphic filtering. Non-skin regions in the area of face were removed by skin recognition method. And different areas of skin were weighted to solve the problem of different reflectance characteristics of skin. Furthermore,de-linear in conjunction with average filter were performed to eliminate the influence of noise.Second, applying canonical correlation analysis(CCA) methods to extract the heart period information from the processed video images. The two methods CCA-basedand ICA-based were compared. The results of correlation and Bland-Altman analysis showed that the algorithm of non-contact heart rate measurement based on the method of CCA has a better performance both in accuracy and cost time.Third, methods of face tracking was utilized to solve motion interference in the process of detection of heart rate. Camshift method was introduced for single face tracking. While for multi-subject, Harris corner feature combined with KLT Tracking method were proposed. An experiment had been carried out in order to verify the methods. Both for single and multi-subject heart rate could be detected in a decent level of accuracy even in the state of motion.In summary, some optimization methods for preprocess of video images and extraction of the period of heart rate were studied in this paper. And method of obtaining reliable heart rate in the case of multi-subject in motion state was proposed.The verification of experiment showed reasonable results has been achieved.