SSAGCN: Social Soft Attention Graph Convolution Network for Pedestrian Trajectory Prediction.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Accurate pedestrian trajectory prediction is crucial for autonomous driving safety.
• Existing models often struggle to simultaneously capture complex social and environmental interactions.

Purpose of the Study:

• To develop a novel model, the Social Soft Attention Graph Convolution Network (SSAGCN), for enhanced pedestrian trajectory prediction.
• To effectively integrate social interactions among pedestrians and environmental influences into the prediction model.

Main Methods:

• Proposed a Social Soft Attention mechanism to model nuanced pedestrian-pedestrian interactions, adapting to varying situational influences.
• Introduced a sequential scene sharing mechanism to propagate environmental context across pedestrians spatially and temporally.
• Utilized graph convolution networks to process interaction information.

Main Results:

• The SSAGCN model successfully generated socially and physically plausible pedestrian trajectories.
• Achieved state-of-the-art performance on public pedestrian trajectory prediction datasets.
• Demonstrated the effectiveness of the proposed attention and scene sharing mechanisms.

Conclusions:

• SSAGCN provides a robust framework for autonomous driving by accurately predicting pedestrian behavior.
• The model's ability to integrate social and environmental cues leads to more realistic trajectory predictions.
• The developed approach advances the field of pedestrian prediction in complex traffic scenarios.