Smartphone Sensor Fusion with Neural PDR Drift Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing indoor positioning methods using Wi-Fi and PDR face challenges such as infrastructure dependency, RSS fluctuations, and magnetic signature ambiguity, leading to inaccurate localization.
Innovation Solution
A sensor fusion framework combining smartphone inertial sensors, magnetic field data, and Wi-Fi measurements using deep learning techniques, including VAEs and neural networks, to generate accurate absolute and displacement predictions, and a Kalman filter for drift correction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If Wi-Fi fingerprinting is used for indoor positioning, then positioning can be performed in enclosed environments, but severe RSS fluctuations lead to localization errors of 2-3 m
Solution Approach 1:
The patent combines multiple positioning methods (Wi-Fi fingerprinting, PDR, and magnetic field-based localization) into a unified system. The Wi-Fi module provides absolute position references, PDR provides continuous relative motion tracking, and magnetic field data provides landmark-based position corrections. This merging of methods compensates for the weaknesses of individual approaches, reducing the localization error from 2-3 m to below 1 m.
Solution Approach 2:
The system implements feedback mechanisms where the predicted position from PDR is continuously corrected using Wi-Fi fingerprint matching and magnetic field landmark recognition. The localization module receives feedback from multiple sensors and adjusts the position estimate accordingly, creating a closed-loop system that maintains accuracy despite RSS fluctuations.
2Speed
If PDR systems are used for indoor positioning, then relative position tracking is provided, but drift occurs over long distances
Solution Approach 1:
The system performs preliminary absolute position determination using Wi-Fi fingerprinting and magnetic field landmarks at key locations before PDR tracking continues. These preliminary position fixes establish reference points that prevent cumulative drift, allowing PDR to maintain high-speed tracking while periodic corrections keep the position accurate over long distances.
3Ease of manufacture
If magnetic field based approach is used, then infrastructure-free positioning is achieved, but magnetic anomalies create ambiguity in localization
Solution Approach 1:
The system uses Wi-Fi access points and known landmark positions as intermediary reference points to disambiguate magnetic field patterns. When magnetic anomalies create ambiguous signatures, the system refers to the database of known landmark positions and Wi-Fi fingerprint locations to resolve the ambiguity, maintaining both the infrastructure-free advantage and positioning accuracy.
4Reliability
If Wi-Fi and Bluetooth sensors are used for indoor positioning, then positioning is achieved in indoor environments, but a large number of access points with large coverage are required
Solution Approach 1:
The system makes existing Wi-Fi access points serve multiple functions: they provide both network connectivity and positioning references. By using the same Wi-Fi infrastructure for both communication and localization, the system eliminates the need for separate positioning infrastructure, reducing overall system complexity while maintaining positioning availability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of predicting a position of a portable electronic device in an indoor environment from sensor data is disclosed. The method comprises generating at least one absolute position prediction, wherein generating each absolute position prediction is based on radio signal data and/or magnetic field data of the portable electronic device, and generating a displacement prediction, wherein generating the displacement prediction comprises generating, by a neural network, the displacement prediction based on inertial sensor data of the portable electronic device. Generating the absolute position prediction may be repeated at a first rate, and generating the displacement prediction may be repeated at a second rate. The method further comprises determining a predicted position by updating a previously predicted position with at least one of the displacement prediction and the at least one absolute position prediction.