Gaussian Mixture Conditional Variational Recurrent Neural Network for Unified Trajectory Imputation and Prediction

Area of Science:

• Computer Science
• Artificial Intelligence
• Robotics

Background:

• Trajectory prediction is crucial for understanding human behavior but is hindered by incomplete observational data due to occlusions or sensor limitations.
• Existing methods often fail when trajectories are not fully observed, limiting their real-world applicability.

Purpose of the Study:

• To develop a unified framework for trajectory imputation and prediction that robustly handles incomplete observational data.
• To introduce the Gaussian Mixture Conditional Variational Recurrent Neural Network (GMC-VRNN) for enhanced spatio-temporal representation learning.

Main Methods:

• The proposed GMC-VRNN framework integrates a Multi-Space Graph Neural Network (MS-GNN) with a Gaussian Mixture Conditional VRNN.
• A Bidirectional Temporal Decay (BTD) module is incorporated to improve spatio-temporal learning under missing data conditions.

Main Results:

• Extensive evaluations on sports datasets demonstrate the effectiveness of GMC-VRNN in jointly performing trajectory imputation and prediction.
• The GMC-VRNN framework significantly outperforms state-of-the-art methods in terms of precision and robustness with incomplete trajectories.

Conclusions:

• The GMC-VRNN provides a robust solution for trajectory prediction and imputation in the presence of missing data.
• This unified approach enhances the reliability of trajectory analysis in complex, real-world environments.