Development of a Modular ROS-Enabled Pedestrian Intention Prediction Architecture for AVs Maneuvering Control

In this work, the problem of predicting a pedestrian’s intention to cross the road is addressed using visual data from a camera. The proposed ROS-based modular architecture consists of four modules: Visual-Perception, Intention Prediction, and Planning and Control Modules. The Visual Perception module is divided into three sub-modules. The pedestrian detection is responsible for detecting the pedestrian and analyzing his motion and looking states. The lane detection is responsible for analyzing the structured environment which helps in the road state classifiers. The third sub-module aims to extract some curvilinear localization states that are essential for the vehicle’s motion planning and control. The intention prediction module captures the pedestrian’s intention to cross the road. A comparative study is conducted between three different data-driven sequential models. Each model is trained on the JAAD dataset and different extracted features from the visual perception module. The proposed GRU model obtained an 86% average f1-score, anticipating the pedestrian’s intention two seconds in advance when the pedestrian is standing, and three seconds in advance when the pedestrian is walking to the crossing area. To control the maneuver of the vehicle, longitudinal velocity and lateral controllers are implemented to control the motion of the vehicle while avoid collision with the pedestrian based on the intention prediction. Finally, this work is verified on a 1:4 scaled real vehicle to ensure the applicability of implementing this work in real hardware.