UMTrajNet: A spatiotemporal uncertainty modelling structure of human movement trajectories under complex indoor areas

Movement uncertainty modeling is essential for reliable human mobility analytics, especially in indoor spaces with denied Global Navigation Satellite System (GNSS) signals and complex human motion modes. However, the performance of current uncertainty modelling approaches is hindered by the dynamic spatial–temporal correlations within the indoor human movement trajectory. To tackle these problems, this paper proposes a novel uncertainty modelling framework UMTrajNet to address indoor human movement reconstruction, movement uncertainty error prediction, and uncertainty region modelling. A spatiotemporal structure is designed by combining the Graph Convolutional Networks (GCN) network and the Long Short-Term Memory (LSTM) network, enabling adaptive learning the spatial and temporal features of human movement. An adaptive Euclidean distance method is employed to generate continuous uncertainty regions for the selected indoor trajectory based on extracted spatiotemporal features. By integrating GCN with LSTM, UMTrajNet effectively captures the complex spatial–temporal dependencies within the trajectory data thereby overcoming the limitations of traditional solely LSTM-based models in spatial context correlation modeling. Real-world experimental results indicate that the presented UMTrajNet significantly outperforms the state-of-the-art baseline models in uncertainty modelling across different datasets.