Indoor Landmark Recognition System Based On Multi-sensor Fusion

Location information is considered as the important sensing information collected by sensors in the Internet of Things.People spend most of their times in the indoor environment,so that the indoor location-based services have a wide prospect.Meanwhile,along with the enhanced sensing and compute function of the portable device,the indoor location-based services was developed one of another.Such as indoor floor plan,and pedestrian dead reckoning which don't need help from other device.Besides the indoor landmark recognize is an important issue of the research indoor location-based services.For example,the research approach based on Crowd Sensing can recognize and location out the potential indoor landmarks,so that it can provide more plump context for the research of indoor map.when user passed the indoor landmark and be recognized by the system,the deviation of the Inertial navigation can be regulate in order to get more accurate location information.Based on the perception that when user passed some indoor landmark,they will behavior some specific trait,this paper design and developed the landmark recognize system based on the multi-sensor fuse.This system could recognize some indoor landmarks such as elevator,stairs,escalators,vending machines,and subway ticket gate.The main idea is that we Multimodal perceptual of user By the motion sensors such as Accelerometers,gyroscopes and magnetometers in the smart mobile phone.Then we extract and recognize the behavior feature when user passed different landmarks,ensure that we can recognize the common indoor landmarks acurrately.This study mainly contains three parts: Data acquisition,data processing and landmark recognition.The first part is data acquisition: it mainly Multidimensional perception the user's behavior.It contains the data acquisition from Accelerometers,gyroscopes and magnetometers.The second part is data processing: in order to get stable user's motion behavior,we will pre-treated the multimodal data.It'll turn the gyroscope and magnetometer coordinates into global coordinates,and use low-pass filter to eliminate noise from accelerometer scale.The third part is landmark recognition: it finished feature extraction,decision-making integration by using data fusion and accomplished landmark recognition by the way of machine learning.From this experiment,we validated the way of landmark is available.