HiBeat: A novel highly accurate implementation of cardiac pulse measurement on a multicore architecture

Abstract

Heart rate measurement plays a major role in diagnosis of heart diseases. There are many existing contact based methods which are in practice. These methods tend to be more expensive and unreachable in emergency scenarios. This paper introduces a novel non-contact method called HiBeat. HiBeat calculates heart rate using facial video of the subject. Proposed method does face recognition, traces the color channels and normalizes them. After normalization HiBeat detrends the color channels and then converts it into independent signals by applying independent component analysis. These signals will be converted into frequency domain and band limited to 1-4Hz. Peak value in band limited frequency among all three channels is considered as source for blood pulse per minute unit conversion. HiBeat is thoroughly tested for its accuracy in comparison with OMRON which is a contact based standard tool for heart rate measurement. It is observed that HiBeat results are accurate. HiBeat achieves 81percent accuracy in comparison with existing non-contact methods. HiBeat is parallelized for multicore architecture and it achieves 2x performance compared to its serial implementation.