Predicting pedestrian crosswalk behavior using Convolutional Neural Networks.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Traffic Safety Engineering

Background:

• Pedestrian accidents are a major cause of traffic casualties.
• Current pedestrian signal systems rely on manual activation, which can be unreliable or impossible for some individuals.
• There is a need for automated solutions to enhance pedestrian safety at crosswalks.

Purpose of the Study:

• To design and evaluate an automated system for detecting pedestrians and cyclists at crosswalks.
• To develop a Convolutional Neural Network (CNN) model for real-time pedestrian and cyclist intention recognition.
• To improve crosswalk safety by automatically triggering pedestrian signals when necessary.

Main Methods:

• A dataset of images was collected to train a CNN model.
• The CNN model was designed to distinguish pedestrians and bicycle riders crossing the street.
• A threshold system was implemented to trigger signals based on prediction confidence, and the system was tested in real-world environments.

Main Results:

• The CNN model achieved an average accuracy of 84.96% in predicting pedestrian and cyclist intentions.
• The system demonstrated a low absence trigger rate of 0.037%.
• Prediction accuracy varied, with pedestrians being detected more accurately than cyclists.

Conclusions:

• The automated crosswalk safety system is feasible as a supplementary measure to existing pedestrian signals.
• The system has the potential to significantly improve overall street crossing safety.
• Further accuracy improvements can be achieved through larger, location-specific datasets and advanced computer vision techniques.