Research On Indoor Location Based On Bluetooth Sensor Network And Application In Activity Recognition

Location technology and activity recognition are popular research fields in pervasive computing, by sensing the information of user's location and activity status, then can service at proper time and space automatically. They are applied widely in indoor and outdoor navigation, target tracking, smart home, message delivering, medical monitoring, social network mining and other fields, and play important roles in smart environment application and personalized services.For Bluetooth technology has been widely applied in daily lives, it meets the requirement of pervasive computing environment. This dissertation focuses on using the Bluetooth technology and machine learning methods to solve the problem of indoor high-precision location and user's activity pattern recognition. The localization accuracy surpasses the former research work by the method of Kernel Ridge Regression to model the Bluetooth signal strength and realize the adaptive calibration-free. According to the relevance between user's activity and location, an activity recognition model has been proposed which infers the generalized activity information of higher-level users from the low-level Bluetooth signal strength. The test verifies the feasibility of proposed framework and method, and it further develops the application research of Bluetooth technology in ubiquitous environment.The major works of this dissertation includes following points:A location method based on Kernel Ridge Regression(KRR) is proposed to solve the problem of indoor high-precision location and the signal strength's dynamic changes, which applies the signal strength and physical location information provided by Bluetooth beacons to construct the regression model of signal strength and physical coordinates, and achieve high precision positioning performance. Using on-line collected signal samples to real-time updates the model parameters, the system adapts environmental dynamic changes with better practicability and robustness.A rule-based activity recognition model mapping from low-level Bluetooth signal strength information to user high-level generalized activity has been established. Based on the information of user's location, the dissertation carries out the research of activity recognition by applying location-based probability model and KNN&SVM classification algorithm to identify and analyze the activity patterns and is able to achieve high recognition rate. We construct a Bluetooth sensor network, including the data collection of low-level Bluetooth signal strength information and preprocessing, system construction designation, network communication and data synchronization acquisition, and systematically analyze the characteristics of the Bluetooth inquiry and the features of system work.We collect experimental data from a real office environment to verify the experiment, the results show that use of low-level Bluetooth signal information and Bluetooth sensor network can achieve high-accuracy indoor location and effective identification of user's activity patterns to provide intelligent services.