Enhancing GNSS Precision for Mobile Devices with Sensor Fusion Techniques: A Case Study on eBike Tracking Using State Estimation
(2023)Department of Automatic Control
- Abstract
- Electric bicycles have over time become a common method of transportation. With a rapidly increasing user base and expensive prices, there is also an increasing demand for safety and insurance. Bike safety can be used to notify the bicycle owner of a potential theft, or to locate the position of the bike when it is lost. Improving the autonomous vehicle’s position tracking when its location is unknown to the owner, simplifies the search for a lost or stolen bike.
With the use of an accelerometer and gyroscope as input, coupled with a Global Navigation Satellite System, a prediction algorithm, or filter, was developed to predict the trajectory. Three different dynamical models for the filter were tested for robustness and optimization of... (More) - Electric bicycles have over time become a common method of transportation. With a rapidly increasing user base and expensive prices, there is also an increasing demand for safety and insurance. Bike safety can be used to notify the bicycle owner of a potential theft, or to locate the position of the bike when it is lost. Improving the autonomous vehicle’s position tracking when its location is unknown to the owner, simplifies the search for a lost or stolen bike.
With the use of an accelerometer and gyroscope as input, coupled with a Global Navigation Satellite System, a prediction algorithm, or filter, was developed to predict the trajectory. Three different dynamical models for the filter were tested for robustness and optimization of filter parameters to yield desired results. An Extended Kalman Filter was used for the predictions, while the tested dynamic models were of linear, first-order and second-order types. The simulations used for the model evaluation utilized four different noise models. While the Inertial Measurement Unit contained known noise, the GNSS noise remained unknown.
When the tests with simulations indicated which models had the best performance, the filter was used on data from real-world measurements. Calibration was made on the Inertial Measurement Units’ inner coordinate frame to get position estimates for comparisons with satellite points.
In some cases, the lone use of a GNSS for position tracking proved to be the best, while in other cases the filter output had a higher accuracy of predicting the position correctly. On average, the second-order model proved to have the best performance, concluding that it also was the most robust model. The model had an average error of 1 meter from the true position at best, and 1.4 meters at worst. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9134651
- author
- Byström, Richard and Sjödin, William
- supervisor
- organization
- year
- 2023
- type
- H3 - Professional qualifications (4 Years - )
- subject
- report number
- TFRT-6212
- other publication id
- 0280-5316
- language
- English
- id
- 9134651
- date added to LUP
- 2023-08-18 15:03:42
- date last changed
- 2023-08-18 15:03:42
@misc{9134651,
abstract = {{Electric bicycles have over time become a common method of transportation. With a rapidly increasing user base and expensive prices, there is also an increasing demand for safety and insurance. Bike safety can be used to notify the bicycle owner of a potential theft, or to locate the position of the bike when it is lost. Improving the autonomous vehicle’s position tracking when its location is unknown to the owner, simplifies the search for a lost or stolen bike.
With the use of an accelerometer and gyroscope as input, coupled with a Global Navigation Satellite System, a prediction algorithm, or filter, was developed to predict the trajectory. Three different dynamical models for the filter were tested for robustness and optimization of filter parameters to yield desired results. An Extended Kalman Filter was used for the predictions, while the tested dynamic models were of linear, first-order and second-order types. The simulations used for the model evaluation utilized four different noise models. While the Inertial Measurement Unit contained known noise, the GNSS noise remained unknown.
When the tests with simulations indicated which models had the best performance, the filter was used on data from real-world measurements. Calibration was made on the Inertial Measurement Units’ inner coordinate frame to get position estimates for comparisons with satellite points.
In some cases, the lone use of a GNSS for position tracking proved to be the best, while in other cases the filter output had a higher accuracy of predicting the position correctly. On average, the second-order model proved to have the best performance, concluding that it also was the most robust model. The model had an average error of 1 meter from the true position at best, and 1.4 meters at worst.}},
author = {{Byström, Richard and Sjödin, William}},
language = {{eng}},
note = {{Student Paper}},
title = {{Enhancing GNSS Precision for Mobile Devices with Sensor Fusion Techniques: A Case Study on eBike Tracking Using State Estimation}},
year = {{2023}},
}