@techreport{oai:ipsj.ixsq.nii.ac.jp:00240597,
 author = {Yuqiao, Wang and Takuya, Maekawa and Yuqiao, Wang and Takuya, Maekawa},
 issue = {22},
 month = {Nov},
 note = {Predicting movement trajectories of individuals within indoor environments is essential for applications such as improving navigation systems and enhancing spatial awareness in smart buildings. Traditional trajectory prediction methods often perform map-matching by leveraging floormap information, resulting in complex and costly systems due to feature extraction from the floormap information and integrating the extracted floormap information into computationally expensive non-linear systems such as particle filters. This study addresses these challenges by introducing a user-friendly, end-to-end model that eliminates the need for complicated pre-processing or specialized data collection. Our approach simplifies the task by using only Inertial Measurement Units (IMU) data and a 2D floorplan image to construct a neural network-based trajectory prediction system. By leveraging Graph Neural Networks (GNNs) to integrate user's positional information into the floor map information and Long Short-Term Memory (LSTM) networks to capture the temporal dynamics of user movements, the model reconstructs the trajectory of the pedestrian based on IMU data. This method offers a practical and accessible solution for accurate trajectory prediction in complex indoor settings by integrating spatial and temporal data into a unified framework, demonstrating the effectiveness of GNNs in processing spatial structures for real-world applications., Predicting movement trajectories of individuals within indoor environments is essential for applications such as improving navigation systems and enhancing spatial awareness in smart buildings. Traditional trajectory prediction methods often perform map-matching by leveraging floormap information, resulting in complex and costly systems due to feature extraction from the floormap information and integrating the extracted floormap information into computationally expensive non-linear systems such as particle filters. This study addresses these challenges by introducing a user-friendly, end-to-end model that eliminates the need for complicated pre-processing or specialized data collection. Our approach simplifies the task by using only Inertial Measurement Units (IMU) data and a 2D floorplan image to construct a neural network-based trajectory prediction system. By leveraging Graph Neural Networks (GNNs) to integrate user's positional information into the floor map information and Long Short-Term Memory (LSTM) networks to capture the temporal dynamics of user movements, the model reconstructs the trajectory of the pedestrian based on IMU data. This method offers a practical and accessible solution for accurate trajectory prediction in complex indoor settings by integrating spatial and temporal data into a unified framework, demonstrating the effectiveness of GNNs in processing spatial structures for real-world applications.},
 title = {Preliminary Investigation of End-to-end Indoor Trajectory Prediction Based on 2D Floorplan Image and Graph Neural Network},
 year = {2024}
}