Adaptive ROI for Collision Warning Mitigation Based on Road Geometry and Kinematics

Abstract

This article presents the adaptive region of interest (ROI), based on simple road geometry and vehicle kinematics, for collision warning (CW) mitigation of vehicles during cornering. When the ROI is directed straight ahead, the collision threat assessment may detect objects that are not causing a collision, as they are not positioned along the actual predicted path of the ego vehicle. Therefore, we suggest the ROI adapted to the vehicle’s motion modeled by its kinematics. For real-world applications, the suggested ROI is designed to depend on readily accessible variables such as the steering angle and longitudinal velocity. Furthermore, the adapted ROI is additionally modified to consider the road geometry of the cornering lane. Given the worst case scenario regarding road geometry, an additional road geometry estimation step becomes unnecessary. In addition, a field-of-view (FOV) conflict check verifying FOV violation of the proposed ROI is suggested to confirm whether the adopted sensor is eligible for the proposed ROI. The proposed methods are validated through real vehicle experiments, the results of which demonstrate that the proposed adaptive ROI 1) enables the vehicle to detect potentially threatening objects in the corner within the ROI, where the performance is increased by 140% and 2) the detection rate of unnecessary object located along the road boundary is decreased by 76%. The demonstration video is provided at the following link: https://youtu.be/tsgI6J421y0?si=vwYDLwy9ApvGT_2a