Indoor Position Estimation Using Acceleration-Based Data Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing position estimation techniques for indoor terminals face accuracy issues due to the inability to acquire suitable training data when the terminal is in use, leading to decreased model performance.
Innovation Solution
A position estimation system that uses a processor and storage device to associate measurement data from a terminal device with a position identifier, specifying a period of user movement based on acceleration or step data, and creates a position estimation model using this data to accurately estimate the terminal's position.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If training data is acquired from terminal during actual user use, then model accuracy can be improved, but the data includes unusable information that decreases model accuracy
Solution Approach 1:
The patent extracts only the useful portion of training data by identifying periods when the terminal is actually stationary at a location. It separates usable data (when terminal is stationary) from unusable data (when terminal is moving), keeping only the relevant segments for model training.
Solution Approach 2:
The patent performs preliminary filtering of training data before model training by using acceleration and step data to identify valid stationary periods. This preliminary action ensures that only high-quality data is used for training, preventing degradation of model accuracy.
2Measurement precision
If position estimation model is created for each terminal, then estimation accuracy is improved, but data processing complexity increases
Solution Approach 1:
The patent creates a dedicated position estimation model for each terminal by copying and customizing the base model with terminal-specific training data. This allows each terminal to have its own optimized model while using the same underlying architecture and training methodology.
Solution Approach 2:
The patent applies local quality by tailoring the training data selection and model parameters to each specific terminal's characteristics and usage patterns. Each terminal receives customized training based on its own stationary period data, improving individual accuracy.
Data Source
AI summary
A position estimation system includes: a processor; and a storage device. The storage device stores data associating measurement data obtained by a terminal device with a position identifier. The measurement data includes on a measured radio wave data, and data a measured acceleration. The processor specifies a period from when the terminal device acquires the position identifier until the processor determines that a user of the terminal device has moved; uses training data including the measurement data obtained by the terminal device during the specified period and the position identifier corresponding to the measurement data obtained by the terminal device to create a position estimation model that outputs the position identifier as a position estimation result when the position estimation model receives the measurement data; and when the processor receives the measurement data, estimates a position of the terminal device by using the measurement data and the position estimation model.