Trajectory Prediction of Human-Driven Vehicles on the Basis of Risk Field Theory and Interaction Multiple Models

Abstract

This study focuses on predicting the motion states and intentions of HDVs at unsignalized intersections. On the basis of a risk field-driven driving behavior model for uncontrolled intersections, multiple motion hypotheses are formulated to characterize the motion planning process of drivers in multivehicle conflict scenarios. Each motion hypothesis is modeled and expressed separately via the extended Kalman filter (EKF) model. These EKF models were combined to construct an interacting multiple model (IMM) framework. This framework estimates which motion hypothesis the driver is more likely to adopt as a strategy. By integrating the predictions of multiple motion hypotheses, more accurate predictions are obtained. Ultimately, it estimates the driver's travel path and acceptable risk level and predicts the spatiotemporal trajectory of HDVs within a future time window.