The Algorithm Of Action Recognition And Heart Rate Detection Based On Acceleration And Pulse Wave Signals

Heart rate is an important physiological parameters of human body, furthermore the information of action can be obtained from the exercise heart rate effectively. Many companies have developed wearable watch or heart rate bracelet recently, such as Huawei and Millet. However the action recognition and exercise heart rate detection algorithms are still in development for these watches. These devices achieve the information of motion state and exercise heart rate through analyzing acceleration signals as well as the pulse wave signals. While the accuracy of the devices during people movement still needs improvement. In view of the current situation, this paper intends to study on action recognition and heart rate detection algorithms.The study designed a heart rate watch which can be used in the experimental conditions, that can collect two channels pulse wave signal and three axis acceleration signals at the same time, this device is the basis for follow-up study. Then we developed a set of action recognition algorithm based on three axis acceleration database in public. This research extracted time domain features, the frequency features and time frequency features of the acceleration signals, then used SVM and decision tree to design the action recognition algorithm. The identify results show that the algorithm can identify rest, walking, running, jumping, marking time effectively. The action recognition rate of decision tree is more than 93% respectively. After that, was design exercise test, we used the watch to collect acceleration signals and made use of these signals to design action recognition flow chart, the results show that the average rate of action recognition is 81.78%.Then we researched the heart rate detection algorithm. Pulse wave is sensitive to the moment of a body. The motion artifacts affect the pulse wave interference greatly, which affect the subsequent analysis results, differentexercises cause different interference to pulse wave. We design experiments to obtain the acceleration signals and pulse wave signals in the same time, found that the two types of signals have the same spectral components through the analysis. From this the paper puts forward power spectrum offset to realize the heart rate detection technology. We use the experiments to validate the performance of the algorithm. The experiment collected 10 groups of quiet state, 22 groups of running state, 21 groups of playing football state. During the experiment the subjects wore the Polar belt, the data collected by polar as the standard of heart rate, we used mean absolute errorE′, the percentage of error which less than 5BPM(PE<5), the percentage of error which less than 10BPM(PE<10) to evaluate the accuracy of heart rate algorithm. The experimental results show that E′were3.80 BPM, 6.94 BPM, 16.38BPM; thePE<10were 94.4%, 83.5%, 47.28%,thePE<5were 76.77%, 68.09%, 32.06%in the three kinds of experiments respectively. The results show that during playing football, power spectrum cancellation effect is not ideal. Then we chose independent component analysis, empirical mode analysis, and adaptive filtering method to analysis the data when the subjects were playing football. By comparison, we selected independent component analysis to eliminate motion artifacts, after this method the accuracy of heart rate is raised higher, after the improved algorithm E′reduces the 1.8 BPM. PE<10and PE<5 were increased by 5.73% and 3.55%, respectively.This paper put forward an action recognition which can obtain the higher accuracy, the heart rate detection algorithm we designed can get better accuracy in quiet state and running state, but then when the subjects were playing football,the accuracy needs to be improved. In summary the algorithm is expected to in the exercise heart rate watch.