Research Of A Smartphone Based Indoor Pedestrian Dead Reckoning System

Location Based Service (LBS) has played a more and more importantrole in military and government industries, emergency services, and thecommercial sector. Unfortunately, the Global Navigation Satellite Systemwhich is dominant in outdoor localization can hardly provide indoorlocalization due to high signal attenuation and severe multipath effect.Nowadays truly ubiquitous smartphones are undoubtedly the main carrierfor personal location based services. The powerful context-sensing abilitymakes smartphone a potential promise for the use of indoor localization.This paper focuses on using smartphone’s MARG (Magnetic, AngularRate, and Gravity) sensors to implement pedestrian dead reckoning whichis then fused which indoor floorplan based on particle filtering. Since ourindoor localization strategy merely depends on user’s commercialsmartphone, no extra infrastructures are required. There a few challengeslies in the smartphone based localization solution including: smartphone’shigh degree of freedom and multiple carrying modes, the lowsampling-rate noisy MARG sensor, serious indoor magnetic fieldinterferes, as well as the intensive computational load in particle filtering.To conquer those challenges, this paper focuses on the3following aspects:1. The magnetic anomalies immune attitude algorithmThe severe indoor magnetic interferences from steel and concreteskeletons of modern building and electric power systems will deflectsensor heading estimation and consequently result in error in pedestrianheading estimation. The author has proposed a two-phase filter todetermine the true sensor attitude. In the first phase, the proposed filterplaces generalized likelihood ratio tests on magnetic modulus, horizontal component and inclination, which are then integrated by fuzzy logic; at thesecond phase, the heading sequence from both gyroscope and magneticmeasurements are compared with each other. Output of the two-phase filteris used to fused sensor attitude which is proved to be immune to magneticanomalies by practical experiments.2. The pedestrian heading algorithmIn reality, a smartphone can be carried by the user in multiple wayswhich are summarized into4modes (compass mode, calling mode, pocketmode and swinging-hand mode) in this paper. To deal with all these modesas well as the high degree-of-freedom when phone is carried by user, twodifferent heading algorithms are proposed. One is to estimate pedestrianheading from sensor yaw at particular interfering point; the other is toestimate heading from the principle component’s direction from horizontalacceleration. The heading algorithm as well as pedestrian dead reckoningperformance is then verified in long distance experiments covering all4modes.3. The adaptive particle filtering algorithmThe computational load of particle filtering is proportional to thenumber of particles used. The proposed adaptive particle filteringalgorithm which has the ability to fix heading error and pedestrian walkingparameters by utilizing the information from last generation of survivingparticles uses as few as500particles to achieve sub-meter localizationaccuracy in indoor experiments. The author has also applied binary particleweight and setup a grid map lookup table to minimize the complexity inparticle weight update.All in all, this paper focuses on challenges in setting up a convenientsmartphone pedestrian dead reckoning based localization solution andenables every user to access the indoor localization application using onlyher smartphone.