Prediction of Intelligent Vehicle-Pedestrian Conflict in a Highly Uncertain Environment

Abstract

Ensuring that intelligent vehicles do not cause fatal collisions remains a persistent challenge due to pedestrians' unpredictable movements and behavior. The potential for risky situations or collisions arising from even minor misunderstandings in vehicle-pedestrian interactions is a cause for great concern. Considerable research has been dedicated to the advancement of predictive models for pedestrian behavior through trajectory prediction, as well as the exploration of the intricate dynamics of vehicle-pedestrian interactions. Such endeavors aim to enhance the ability of intelligent vehicles to respond appropriately to the presence of pedestrians in their surroundings. Predicting pedestrian movements in crowded spaces is complex due to pedestrian interactions' continuous and forward-looking nature. While current methods fail to account for temporal correlations among these interactions, recurrent neural networks have been used to model social interactions. However, they are limited in capturing spatiotemporal interactions. Graph Neural Networks (GNNs) have been introduced to address this limitation but do not fully capture real-world social interactions as they consider the impact between traffic participants as fixed or symmetric. We propose a novel graph-based trajectory prediction model for vehicle-pedestrian interactions called Holistic Spatio-Temporal Graph Attention Trajectory Prediction (HSTGA) to address these limitations. HSTGA first extracts vehicle-pedestrian interaction spatial features using a multi-layer perceptron (MLP) sub-network and max pooling. Then, the vehicle-pedestrian interaction features are aggregated with the spatial features of pedestrians and vehicles to be fed into the LSTM. The LSTM is modified to learn the vehicle-pedestrian interaction adaptively. Moreover, HSTGA models temporal interactions using an additional LSTM. Then, it models the spatial interactions among pedestrians and between pedestrians and vehicles using Graph Attention Networks (GATs) to combine the hidden states of the LSTMs. Then, the predicted trajectories of vehicles and pedestrians are used to design a vehicle-pedestrian conflict model. The conflict model is used to investigate safety-critical driving metrics, such as severity and near-miss metrics, as well as normal driving metrics, including vehicle speed, vehicle acceleration, vehicle heading angle, pedestrian speed, pedestrian acceleration, pedestrian heading angle, and distance distributions. Two safety indicators, namely Time-to-Collision (TTC) and Post-Encroachment Time (PET), are used to quantify the severity and near-miss conflicts between the vehicle and the pedestrian. We evaluate the performance of HSTGA on three different scenario datasets, including complex unsignalized roundabouts with no crosswalks and unsignalized intersections. Results show that HSTGA outperforms several state-of-the-art methods in predicting linear, curvilinear, and piece-wise linear paths of vehicles and pedestrians and can also predict collisions between pedestrians and vehicles several seconds before they occur. Our approach provides a more comprehensive understanding of social interactions, enabling more accurate trajectory prediction for safe vehicle navigation.