Predicting Pedestrian Crossing Intention at Signalized Intersections Using Roadside LiDAR Sensors

Detecting, tracking, and predicting pedestrians crossing intention as they approach an intersection could increase the safety and efficiency of signalized intersections. LiDAR (Light Detection And Ranging) traffic sensors, capable of detecting, classifying and tracking objects, could play an essential role in multimodal intelligent infrastructure. In particular, they could complement or even substitute existing traffic sensors such as inductive loops or pedestrian push-buttons. In this paper, we present the results of a case study in a real signalized intersection in Germany, where pedestrian trajectories were collected using roadside LiDAR sensors. Then, we processed the trajectories and conceptualized and prototyped different models to predict pedestrian crossing intention using LiDAR data. This is the first time such models are used for signalized intersections in the literature. Our results demonstrate that already-published formulations can be successfully adapted to signalized intersections, opening up new possibilities for traffic signal control at intersections and Vehicle-to-Everything (V2X) applications. The four prototype models --random forest, logistic regression, classification tree, and artificial neural network-- achieved accuracies near 90%, and better results could be obtained by applying more complex techniques (e.g., using deep learning) and improving input tracking data. In the future, the prediction generated with our models could be communicated to nearby vehicles via an in-vehicle information message or used to activate warning signals. Furthermore, it could be employed to trigger pedestrian prioritization traffic signal control measures.     «

Detecting, tracking, and predicting pedestrians crossing intention as they approach an intersection could increase the safety and efficiency of signalized intersections. LiDAR (Light Detection And Ranging) traffic sensors, capable of detecting, classifying and tracking objects, could play an essential role in multimodal intelligent infrastructure. In particular, they could complement or even substitute existing traffic sensors such as inductive loops or pedestrian push-buttons. In this paper, w...     »