Bimodal Extended Kalman Filter-Based Pedestrian Trajectory Prediction.

Area of Science:

• Computer Science
• Robotics
• Artificial Intelligence

Background:

• Pedestrian trajectory prediction is crucial for autonomous systems and crowd management.
• Existing methods often struggle to capture the dual-state nature (moving/stationary) of pedestrian behavior.
• Accurate prediction requires accounting for social interactions and environmental constraints.

Purpose of the Study:

• To develop a pedestrian trajectory prediction algorithm leveraging the dual-state nature of pedestrian movement.
• To enhance prediction accuracy by incorporating social interactions and physical obstacles.
• To create a comprehensible model with a limited number of parameters.

Main Methods:

• Utilized a bimodal extended Kalman filter (BEKF) to model pedestrian movement.
• Applied a dual-mode probability model to represent pedestrian states (moving or stationary).
• Estimated pedestrian state distributions and predicted future trajectories based on posterior probabilities.

Main Results:

• The BEKF effectively estimates individual pedestrian state distributions.
• The algorithm successfully predicts future trajectories for all individuals in a scene.
• The model achieves high accuracy comparable to deep learning models with fewer than fifty parameters.

Conclusions:

• The proposed bimodal extended Kalman filter offers an effective and comprehensible approach to pedestrian trajectory prediction.
• The algorithm's ability to consider social dynamics and environmental factors enhances its practical applicability.
• This method provides a robust alternative to complex deep learning models for trajectory forecasting.