Research Of Activity Recognition And State Monitoring Methods Of The Human Based On Wearable Device

Human activity recognition and state monitoring is an emerging research hot spot in the field of artificial intelligence and pattern recognition. In medical monitoring, intelligent home, aged care and other fields have a wide range of applications. Combined with the development of sensor technology, low power consumption, wireless physiological sensor network makes the micro sensor network to our daily life environment, the development of wearable small smart device of the difficulty and the cost is greatly reduced. In the field of mobile health, wearable devices have become an indispensable part. On the basis of the above, this paper makes a systematic study on the methods of human activity recognition and state monitoring based on wearable devices.Activity recognition is the basis for the state monitoring of the human body. First, we give the overall scheme of activity recognition, and determine the scope of activity recognition. Considering the complexity of the data and monitoring scheme to the flexibility, this paper collect human activity data by accelerometer, altimeter, heart rate sensor and sensor is placed on the wrist. Because of the acceleration sensor of high sampling frequency and more noise, we use the smoothing filtering and data segmentation in acceleration signal processing, and extract fragment data in time domain and frequency domain feature, construct the feature vector set of examples for activity recognition.Based on the activity of the overall recognition scheme and the set of feature vector samples, we used the decision tree, random forest, artificial neural network constructed classifier, verify the correctness of the classifier and classifier is defined as the activity model. Then, the classifier is compared from three aspects: time complexity of model training, time complexity of the data classification and the accuracy of the classification. Finally, the recognition rate of walking, upstairs and rest reached more than 90%, the recognition rate of running and walking down stairs reached more than 80%, the recognition rate of jump reached 75%.Based on the activity recognition results, we realize the exercise and sleep monitoring. We use pedometer to realize exercise monitoring that based on walk, upstairs, downstairs, running that provides a data reference for user. Based on the results of static recognition, we proposed a three-stage recognition method of sleep time statistics, pedometer had a margin of error of less than 15%, sleep monitoring can be more accurate statistics of the starting and ending time of sleep.