University of Twente Student Theses
Developing an end-to-end Indoor Localisation Scheme using Deep Learning Models and Inertial Odometry Techniques
Das, Subhraneil (2024) Developing an end-to-end Indoor Localisation Scheme using Deep Learning Models and Inertial Odometry Techniques.
Full text not available from this repository.
| Full Text Status: | Access to this publication is restricted |
| Abstract: | Smartphone-based indoor localization has attracted considerable attention in both research and industrial areas. However, the localization accuracy and robustness are still challenging problems due to the influence of noise and drift in the commercially available MEMS sensors housed in the smartphones, the unavailability of GPS, especially in complicated indoor environments. Utilising the sensors widely available across smart-phone devices this thesis report aims to answer that whether an end-to-end solution from sensing to the prediction of trajectories can be developed involving inertial odometry techniques and aspects of deep learning models. A baseline model, i.e., the Robust Neural Inertial Navigation model (RoNIN) is employed to study the mechanism and flow of a data-driven indoor localization scheme for trajectory and heading estimation. The study leads us to formulation of the essential research questions. The thesis proceeds with the study and implementation of the the Rotation- equivariance-supervised Inertial Odometry (RIO) algorithm for the prediction task as mentioned above. The reason behind modifying the existing baseline model is to incorporate the RIO algorithm which claims to improve trajectory estimations as compared to RoNIN, based on combining an auxiliary loss calculation and the MSE loss criterion. This is done primarily to cover corner cases of natural human motion where the velocity of the subject is not zero but less than normal walking velocity. In addition to this, the algorithm proposes the use of uncertainty estimations to decide on conditional updation of model parameters that in turn ensure that the predictions are more accurate. This paper aims to investigate, develop and evaluate the RIO algorithm followed by testing it using existing datasets and freshly collected data. The results from the models are then discussed with respect to the evaluation metrics and is followed by relevant discussion and possible future improvements. |
| Item Type: | Essay (Master) |
| Clients: | ALTEN NL, Apeldoorn, Netherlands |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 50 technical science in general, 54 computer science |
| Programme: | Embedded Systems MSc (60331) |
| Link to this item: | https://purl.utwente.nl/essays/103519 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page