Study On Indoor Floor Recognition And Height Estimation Based On WiFi/BLE/HAR

In indoor environment,location-based services(LBS)is inseparable from indoor positioning technology.In today's society,high-rise buildings are distributed everywhere,3D indoor positioning technologies have become more important.Due to widespread utilization of smartphones and wireless signals such as Wireless-Fidelity(WiFi)and Bluetooth Low Energy(BLE),Pedestrian Dead Reckoning(PDR)and WiFi/BLE positioning technologies have become the most basic and common indoor positioning methods.The associated research has become one of the most crucial positioning problems all over the world.Indoor floor recognition and height estimation based on wireless signals and smart phone inertial sensors have great application value.However,different layouts,variable structures,and sparse wireless Access Points(AP)distribution densities in real multi-storey buildings make the spatial propagation characteristics of wireless signals quite different.Most of floor positioning methods based on wireless signals have certain requirements,so these are difficult to be universal,positioning accuracy and real-time.Although there are many floor recognition or height estimation methods based on air pressure,they also have some shortcomings.For example,the air pressure is sensitive to the variation of the temperature and humidity;the method based on the depended base station requires additional equipment and development of data transmission functions;different smartphones have different barometer models or lack of barometers,etc.These problems have affected the widespread application of the air pressure method.Aiming at the above problems,this research conducts an in-depth study on floor recognition and height estimation based on WiFi/BLE wireless signals and the Human Activity Recognition(HAR)technique.The research consists of the initial/real-time floor recognition method for the wireless signal of two kinds of floor structures;the HAR algorithm based on smart phone three-axis acceleration sensor data;the machine learning classification algorithm and feature vector selection;the floor change detection scheme and height estimation method based on HAR and so on.The main findings are summarized as follows:(1)Based on the theory of the wireless signal floor recognition and consideration of its weak applicability on the indoor space structure,the structure of the multi-floor indoor space is divided into the full-floor floor structure and the atrium space structure.And the spatial distribution characteristics of wireless signals in the two spatial structures are deeply analyzed.(2)For the full-floor structure inside multiple floors,the low universality of AP deployment conditions and environmental changes based on most wireless signal floor recognition methods widely exist.Therefore,a wireless signal interval confidence floor recognition algorithm based on optimization theory,probability,and statistics theories is proposed.Then,a large and complex multi-floor environment with an unknown and uneven AP layout is chosen,and the experiment is carried out.Compared with the majority voting committees,K-means,Naive Bayes and KNN methods,the floor recognition accuracy of the proposed algorithm is identified as the best approach,and both the fingerprint collection workload and the fingerprint database data volume are the smallest.(3)To improve the low accuracy of wireless signal floor recognition methods in the multi-storey buildings with an atrium space,an adaptive weighted fusion floor recognition algorithm based on wireless signals is developed.Three test sites are selected and tested by using two motion types of five seconds of fixed-point static and one second of real-time dynamics.Compared with classification algorithms such as Decision Tree,Support Vector Machine(SVM),KNN,and neural network,the results show that the average accuracy to the floor recognition of the proposed algorithm is base in three test sites and two motion types.(4)Aiming at the problems of low universality of the air pressure method or insufficient real-time detection of floor changes based on the existing method,a floor change detection scheme depending on the combination of HAR and floor landmark parameters is established,in which HAR uses smartphone sensor data to realize with SVM classification algorithm.The verification is conducted by using the movement process of pedestrians to switch multiple floors,and the comparison with the floor recognition method based on wireless signals and air pressure proves the high precision,real-time and stability of the proposed scheme.(5)The wireless signal floor recognition method induced by the multipath effect is not stable enough,and the floor change detection method by HAR cannot be realized independently.Therefore,in the environment of complex multi-floor spatial structure,a fusion scheme of two methods is proposed,and a multi-method fusion experiment is performed to verify the stability,precision and universality of floor recognition of the fusion scheme.(6)To address the insensitivity of smartphone barometer to sub-meter height and associated dependence of air pressure on the humidity,a sub-meter indoor height estimation method based on HAR is proposed.This method uses the process data of the floor change detection,the ladder library and the HAR results to achieve a highprecision height estimation.In the test process of pedestrians going up and down in a multi-storey building,the sub-meter-level height estimation accuracy is achieved in the17-meter test field.Besides,the method has no error accumulation and is independent of environmental impacts as well as good robustness.There are 63 figures,13 tables,179 references in the dissertation.