3D-LiDAR-based Pedestrian Detection for Demand-Oriented Traffic Light Control

Abstract

Traffic lights typically offer push buttons for pedestrians to request crossing the road. Although this method is effective and simple, it has several drawbacks: (1) it requires an explicit user interaction, (2) no information about the number of pedestrians is obtained and (3) there is no information about pedestrians’ vulnerabilities. Thus, it is not possible to optimize the traffic light control in a demand-oriented way. To address this problem, we present a concept for a demand-oriented traffic light control which is based on a novel pedestrian detection and vulnerability classification method. This approach combines privacy-preserving 3D-LiDAR data acquisition and state-of-the-art deep learning methods. An evaluation on real traffic data obtained from two cities in Germany reveals an overall accuracy of 96 % for pedestrian detection and vulnerability classification. Finally, we show how our demand-oriented traffic light control contributes to (1) an automation of pedestrian signal requests, (2) a reduction of pedestrians’ waiting times and (3) an adaption of the green phase’s length according to vulnerabilities.