A Review of Pedestrian Trajectory Prediction Methods Based on Deep Learning Technology

Area of Science:

• Computer Science
• Artificial Intelligence
• Robotics

Background:

• Pedestrian trajectory prediction is crucial for autonomous driving and intelligent urban systems.
• Deep learning models have surpassed traditional methods in handling complex behaviors and social interactions.

Purpose of the Study:

• To systematically review and critically analyze deep learning-based pedestrian trajectory prediction approaches.
• To provide a structured examination of Recurrent Neural Networks (RNNs), Generative Adversarial Networks (GANs), Graph Convolutional Networks (GCNs), and Transformer models.
• To offer a comparative analytical framework for evaluating these methods.

Main Methods:

• Systematic literature review of deep learning models for pedestrian trajectory prediction.
• Analysis of four key model families: RNNs, GANs, GCNs, and Transformer.
• Development of a comparative framework evaluating strengths and limitations against standardized criteria.
• Comprehensive taxonomy of datasets and evaluation metrics.

Main Results:

• Deep learning models demonstrate superior performance in multi-modal behavior and social interaction prediction.
• Comparative analysis reveals distinct strengths and weaknesses across RNNs, GANs, GCNs, and Transformers.
• Identification of established practices and emerging trends in datasets and evaluation metrics.

Conclusions:

• Deep learning is the dominant approach for pedestrian trajectory prediction.
• Future research should focus on semantic scene understanding, model transferability, and the precision-efficiency trade-off.
• This survey provides a historical perspective and guides future research directions.