Research On Motion Robust Video-Based Heart Rate Detection With Blind Source Separation

Heart rate(HR)is an important parameter that can reflect the physical and psychological conditions of an individual.Accurate measurement of heart rate is widely used in medical diagnosis,health monitoring,fatigue driving detection and other fields.In recent years,remote photoplethysmography(r PPG)technology based on ordinary cameras has become a research hotspot.It has lots of advantages such as no contact,low cost,simplicity,convenience,and wide application.However,due to the weak amplitudes of the pulse wave signals indicating physiological status via cameras,the r PPG technique is susceptible to motion interference,resulting in a decreasing accuracy of HR measurement.This paper proposes a motion robust video heart rate detection method based on blind source separation to improve the accuracy and robustness of video heart rate detection.On the one hand,a single camera cannot detect the subject’s face every moment when the subject’s head has a large range of motion or when the subject is walking around in the monitoring environment.This paper realizes a simulation study of remote heart rate detection which use multiple synchronized cameras to seamlessly photograph subjects by combines ensemble empirical mode decomposition(ensemble Empirical mode decomposition,EEMD)and time-delay Canonical correlation analysis(TDCCA).For each frame of image,select the region of interest with the largest area from multiple cameras as the target region of interest.Second,calculate the average value of all pixels in the target area of interest,and form a multi-channel RGB pixel average sequence frame by frame.Then,combined ensemble Empirical mode decomposition and timedelay canonical correlation analysis to obtain the blood volume pulse signal.Then calculate the heart rate.The performance of this algorithm was verified on a selfcollected database and compared with the performance of other four typical methods.The results show that the method obtains the best performance.The average absolute error is 4.11 times/minute,the average percentage error is 5.26%,and the root mean square error is 5.37 times/minute,and the Pearson correlation coefficient is 0.90.On the other hand,a motion interference removal scheme based on the combination of blind source separation and spectral subtraction technology is proposed for the situation where the subject has large head movement.The motion noise source is extracted by joint analysis of the RGB pixel mean sequence of the facial area of interest and the motion trajectory of the facial feature points.And the motion noise source is eliminated from the RGB pixel mean sequence,and then the motion noise is further suppressed by the spectral subtraction technique.Thereby improving the accuracy of video heart rate detection.The public database UBFC-RPPG is used to verify the performance of the algorithm and compare it with the performance of other methods.The results show that our proposed method has the best performance.In summary,this paper provides a solution for video heart rate detection in real environments.