Research On Sleep Recognition Based On Inertial Sensor

In the present and future, the sleeping identification technology has a great practicability and application in people’s life. In the monitoring of sleeping, through the activity of body in human’s sleeping by the long time monitoring, we can evaluate the quality of person’s sleeping, and at the same time it can be used as an auxiliary tool to diagnose some sleep disorders. In addition, with the development of the smart home, "No interaction is the best interaction" is a direction of the future development of smart home. The user’s habits and customs must be constantly learned by smart home control system, which is intelligent control of human-oriented. On the basis of it, the personalized, intelligent smart home control scheme can be designed. As long as the person’s activity be real-time monitored and the person’s state be classified, the equipment in the home can be intelligent controlled by the smart home system according to the specific state of user. The humanized program control can be achieved, and at the same time we can achieve the goal of saving resource.Now, there are a lot of sleep monitoring device based on wearable devices, intelligent hand ring, wrist movement, leg movement. Most of them take the acceleration sensor as the detecting device. The feature of the acceleration sensor be extracted. The activity count be calculated by the comparing be between feature and the setting threshold value. Then the activity count be simply weighting to calculate the evaluation of certain period of time. The evaluation be used as the basis of identifying user’s sleep or wake up. This method is simple and efficient, but the recognition accuracy in some special scene is unsatisfactory. We summed up some feature extraction method of inertial sensor data, and more characteristics combination method which can describe the original signal and classification method be used in the study of sleeping identification in this paper.In this paper, according to the daily life scene, we defined four types of activities, sleep, low intensity, moderate intensity and high intensity. We use three algorithms to classify the activities. Firstly, we collected the data sets by inertial sensor wearing on the wrist. The set be filtering and normalized processing. Then we extracted the feature of mean, variance, peak value and wavelet coefficients. The classification is SVM, LS-SVM and BP neural network. The experiments show that the accuracy of SVM and LS-SVM is higher, and the average accuracy rate more than 90%. The BP neural network classification accuracy is low, only about 76%. Finally, through the voting mechanism of three classification methods, the final label can be calculated. On the basis of the classification results, the sleeping recognition can be finally realized by improvement sleep algorithm.