Forward Collision Avoidance using a Monocular Camera based on ROS2

The automotive industry is rapidly transitioning from hardware-centric to software-centric vehicular architecture, Software-Defined Vehicles (SDVs) to address existing complex $E / E$ architectures. This shift to SDVs will accelerate the evolution from Advanced Driver Assist Systems (ADAS) to autonomous driving by decoupling vehicle software and hardware. Forward Collision Avoidance (FCA) systems are an integral part of ADAS which is closely related to user safety. In this paper, we propose an FCA system implemented using only a monocular camera by utilizing a ROS2-based model car. This system is composed of three parts: Detection, Depth Estimation, and Postprocessing part. Through a monocular camera input, a detection model detects a car in the image while a depth estimation model generates an estimated depth map. And then, they are all combined by multi-thread processing in the postprocessing part. The FCA structural approach utilizes both data to make appropriate driving decisions based on the detected car in front of an ego car. This research is implemented on a ROS2-based model car to ensure real-time performance, which can be extended to autonomous robots and vehicles.