Ped-CrossSafe System: Real-time Monitoring and Crossing Prediction of Pedestrians

Abstract

The University of Central Florida invention introduces Ped-CrossSafe: an intelligent, real-time system designed to improve pedestrian safety at intersections by predicting crossing intentions using advanced AI and computer vision. Traditional systems rely on push-buttons and static signal timing, often leading to unsafe crossings and traffic inefficiencies. Imagine a crosswalk that knows when you're about to cross—and makes traffic stop for you before you even take the first step. This system eliminates those limitations by detecting and tracking pedestrians - including vulnerable users like children, cyclists, and wheelchair users- and proactively activating pedestrian signals. It also extends crossing time when needed, ensuring safer and more inclusive mobility. The result is a smarter, safer, and more responsive intersection experience. 

Technical Details: Ped-CrossSafe uses video feeds from four cameras placed at key points around an intersection to detect and track pedestrians in real time. The system applies deep learning models for pedestrian detection and pose estimation, and advanced object tracking algorithms for tracking movement across defined zones—waiting, start-crossing, and crossing. It calculates features like walking speed, body orientation, and head pose, which are then analyzed using an ensemble machine learning model to predict crossing intentions and direction.

To ensure real-time performance, the system activates pose estimation only when pedestrians are in the waiting area, reducing computational load. It also includes a phase extension checker that monitors pedestrian progress and extends signal time if needed, based on walking speed and crosswalk distance. Optimizations such as reduced video resolution and efficient Python libraries enable processing speeds up to 33 FPS, making the system both fast and scalable for deployment in real-world traffic environments.

Benefit

Market Application