Study On Wearable Motion Recognition And Heart Rate Monitoring

With the advent of the aging era,how to monitor the elderly’s Activity of Daily Living(ADL)and physiological parameters,to better protect the daily health of the elderly,is becoming more and more important.The system adopts STM32+MTK dual-processor system,uses the nine-axis inertial sensor MPU9250 and heart rate sensor MAX30102 to design the data acquisition module,positioning module,low-power module,motion recognition module,heart rate acquisition module software and hardware system.In order to solve the problem of short battery life of wearable devices,a low power consumption algorithm for motion static automatic switching is proposed.The test results show that the current of low power consumption module is reduced by about 90% on average.Aiming at the problems of low recognition rate and low real-time performance in human motion recognition,a motion state recognition algorithm based on time difference and threshold is proposed to identify standing,sitting,walking,running,walking stairs and falling actions.The recognition algorithm adopts the idea of using time segment identification and "one ring set and one ring" for each action,and it is considered valid only when the corresponding action is completed within the corresponding time difference For the sit-up and stand-up phase movements and the continuous movements of walking and running,the feature of static is used to reduce false positives.The experimental results show that the accuracy of the algorithm for each action is above 90%,and the average the false positive rate was 4.33% and the average false negative rate was 2.56%.The classification algorithm of random forest is used to classify the motion state.The variable-scale sliding window segmentation technique is used to extract 27 features of the time-frequency domain of the combined acceleration,combined angular,elevation and roll angles.Each action extracts 108 features and uses the PCA algorithm for feature extraction.Tests show that the accuracy of the fall classification is 100%,and the accuracy of the ADL classification is 93.48%.Compared with naive Bayes,decision tree,and K-nearest neighbor classification algorithm,ADL achieved 86.96%,89.13%,and 91.3% accuracy respectively,and the fall to the action achieved 87.5%,81.25%,and 81.25% accuracy,respectively.It can be seen that the random forest classification algorithm has achieved better results for motion recognition.The heart rate can reflect a person’s health.This article uses the MAX30102 heart rate sensor to collect the heart rate of the human body using the PPG method.The error rate is ±2% compared with the ECG recorder.