Presentation is loading. Please wait.

Presentation is loading. Please wait.

Detection of Magnetic Anomaly in Heading Estimation for Smartphone-Based Indoor Localization Marzieh Jalal Abadi University of New South Wales , Sydney,

Similar presentations


Presentation on theme: "Detection of Magnetic Anomaly in Heading Estimation for Smartphone-Based Indoor Localization Marzieh Jalal Abadi University of New South Wales , Sydney,"— Presentation transcript:

1 Detection of Magnetic Anomaly in Heading Estimation for Smartphone-Based Indoor Localization Marzieh Jalal Abadi University of New South Wales , Sydney, Australia Thanks for attending the presentation

2 Introduction Pedestrian dead reckoning(PDR): Independent Positioning System It estimate current position with respect to the previous position by monitoring the person’s movement via sensors Magnetometer sensor Perturbation

3 Motivation The proposed hypothesis : Collaborative approach
Need to detect the erroneous heading estimates Exclude them before fusion

4 Nature of Heading Error (using Magnetometer)
Very high heading estimation error(180°) Different heading error at the same location but different direction Heading error is high at certain pocket of space 8.1m × 6.3m Room 264 points in a grid : 45cm Samsung Galaxy SIII Heading error for HT =280° and HT =100°

5 Discuss the pattern and results
Data Analysis Data Collection Data Display Verifying Data Measurement Model Fit Discuss the pattern and results Express Data Generate Data Characterizing Source of Error in data Multivariable analysis Statistical tests: Dependency of multi pedestrian’s heading Factor analysis Conjoint analysis Correlation Parametric and non-parametric Normal Kernel distribution Compare Fit Parameters

6 Data Analysis The magnetometer measurements (heading estimates) can not be modeled with a parametric distribution Perturbation is highly localized and we need a prior knowledge of the magnetic field distribution in each corridor (for different directions).

7 Perturbation Detector
Data Collection: (mx ,my ,mz ) Data Processing 1. (mx , my , mz , Ferr , Herr , Ierr) 2. Time window : 0.5 sec ~ 1 step 3. Binary class : C0 ,C1 4. |hest-hT|<threshold(γ) Training Feature Selection: Correlation-Based Feature Selection Classifier : Logistic Regression, 10cv

8 Perturbation Detector
“Mean” is the selected feature for all parameters Logistic regression can classify local perturbation within 15° error threshold : accuracy 98% in average To model the perturbation in a path, we need a prior knowledge of its magnetic field distribution We can not build a general model that holds for all environments The model of one corridor is not applicable for other corridors One limitation of our method is its dependence on prior model for perturbation detector: Find a general model for perturbation in all environments.

9 Thank you


Download ppt "Detection of Magnetic Anomaly in Heading Estimation for Smartphone-Based Indoor Localization Marzieh Jalal Abadi University of New South Wales , Sydney,"

Similar presentations


Ads by Google