Research On Indoor Positioning Method Based On Deep Learning Magnetic-Field/Visual Feature Fusion

With the development of the mobile Internet,smart devices such as smart phones have been widely used in our daily life.Location based services with mobile terminals have also become an integral part of mobile Internet services.The smart terminal is closely related to the activities of people,and the indoor positioning of the smart terminal becomes an important part of location based services.Most of the current smart terminals contain geomagnetic sensors and cameras.These two sensors are basis for intelligent terminals to provide indoor positioning services without external infrastructures.The main content of this research is the positioning method according to magnetic field characteristics and geomagnetic/visual fusion characteristics in indoor environments.In the research process of indoor positioning methods based on magnetic field characteristics,we discover the stability characteristics of indoor magnetic field in the spatial continuous region and propose indoor magnetic field sequence model.Based on this model,the paper reveals two magnetic field sequence localization algorithms,which effectively realize the indoor positioning of intelligent terminals on the basis of magnetic field characteristics.Due to small amount of geomagnetic feature information,in order to achieve a more robust positioning method,the paper combines the visual features and puts forward a sensor fusion positioning method based on a neural network algorithm.The main contributions of the paper include four points: 1.A geomagnetic sequence matching indoor positioning method according to hidden Markov model is proposed to realize indoor positioning in narrow environments such as offices and corridors.2.Based on the characteristics of geomagnetic sequence extracted by the recurrent neural network,we design a complete neural network model to achieve the effect of geomagnetic sequence localization in both the open and narrow environments.3.In the paper,the convolutional neural network is combined with the Recurrent neural network to extract the features of visual information in the pre-trained neural network model,and the recurrent neural network is designed to extract the information of the image sequence in time dimension.4.On the basis of the features of geomagnetic sequence and the characteristics of visual sequences,this paper proposes a method for feature fusion and localization using a neural network algorithm.In summary,we carry out in-depth research on indoor positioning method based on magnetic field sequence features and indoor positioning method on the basis of magnetic field and vision fusion,and realize indoor positioning system in multiple modes in the paper.