Research on a Pedestrian Crossing Intention Recognition Model Based on Natural Observation Data

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Accurate pedestrian intention recognition is crucial for the safety and efficiency of autonomous vehicles in urban environments.
• Existing methods require enhancement for reliable real-time prediction.

Purpose of the Study:

• To develop and validate an advanced model for recognizing pedestrian crossing intentions.
• To improve the predictive accuracy of pedestrian behavior for autonomous driving systems.

Main Methods:

• Collected natural crossing data using laser scanners and HD cameras.
• Selected 1980 effective pedestrian crossing samples for analysis.
• Developed an attention-based long short-term memory (AT-LSTM) network model.

Main Results:

• The AT-LSTM model achieved 96.15% accuracy in recognizing pedestrian intent 0s prior to crossing, outperforming SVM by 6.07%.
• At 0.6s prior, AT-LSTM accuracy was 90.68%, exceeding SVM by 4.85%.
• Identified key parameters influencing pedestrian crossing intention through statistical analysis.

Conclusions:

• The proposed AT-LSTM model offers superior accuracy for pedestrian intention recognition compared to traditional methods.
• This research provides a foundational framework for enhancing the decision-making capabilities of future automated vehicles.
• Accurate intention recognition is vital for safe human-robot interaction in urban autonomous navigation.