EGAT: Extended Graph Attention Network for Pedestrian Trajectory Prediction.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Pedestrian trajectory prediction is crucial for autonomous driving, robot interaction, and safety monitoring.
• Existing methods often overlook the influence of distant pedestrians, limiting predictive accuracy.
• Limited receptive fields in current models restrict the scope of considered interactions.

Purpose of the Study:

• To develop an advanced model for more accurate pedestrian trajectory prediction.
• To enhance the understanding of pedestrian interactions by incorporating both local and distant agents.
• To improve the foresight and judgment capabilities in applications like autonomous driving.

Main Methods:

• Proposed an Extended Graph Attention Network (EGAT) to expand the receptive field.
• Incorporated temporal models like TSG-LSTM (TS-LSTM and TG-LSTM) and P-LSTM.
• Utilized residual connections within LSTM-based models to improve information transmission.

Main Results:

• The EGAT model demonstrated superior performance compared to state-of-the-art methods.
• Achieved excellent results on the ETH and UCY public datasets.
• Generated more reliable and accurate pedestrian trajectories.

Conclusions:

• The proposed EGAT model effectively addresses limitations of existing methods by considering a wider range of pedestrian interactions.
• EGAT significantly improves pedestrian trajectory prediction accuracy and reliability.
• The enhanced model contributes to safer and more robust autonomous systems.