Distant pulse measurement system for real-time surveillance applications

Touchless heart-rate monitoring is desirable due to hygienic and discreet operation. Several approaches were proposed so far. Our design makes a step towards camera-based cardiac surveillance system. We have developed, and tested a real-time video processing algorithm. Viola-Jones body part detection framework was used for object detection. Than a Region Of Interest was determined for computation of the mean value of the selected components. A sequence of consecutive raw samples y(t) constitutes a buffer subjected to normalization which is followed by noise reduction and limiting the band by means of a temporal FIR and FFT transformation. The pulse rate was detected as the frequency corresponding to maximum energy band in spectral domain. MATLAB was used for designing and testing the system and generated C++ code was compiled to a 32-bit floating point processor. We recorded footage from 2 volunteers at resolution 800×600 and data rate up to 150Hz. Wearable pulse oximeter was used for measuring reference signal. The error of estimated pulse of subjects under controlled illumination and during their physical activities ranged from 3.1 to 18.8 bmp. All computations meet demands of 60fps real time surveillance system, making the pulse-dedicated intelligent camera a handy tool for monitoring humans, their health status and emotions.