Constructing Obstacle Maps for Highly Automated Vehicles Using the Extended Kalman Filter

The paper deals with the development of methods and tools for building obstacle maps for highly automated vehicles (HAV) based on the Extended Kalman Filter (EKF). Obstacle maps are used in the HAV control system for avoiding moving and stationary targets and for localization in conjunction with other positioning tools. The EKF module is used to estimate the position of landmarks and the vehicle itself based on the data received from the front cameras. The data processing involves converting the images into Birds-Eye-View (BEV) projections using an Inverted Perspective Map (IPM) to account for spatial information. The paper describes the methods and algorithms used for image processing, object recognition using a neural network with YOLOv8 architecture, and determination of free space for movement using semantic segmentation based on MobileNetV3 neural network architecture. The results of testing the developed methods and software in the Webots modeling environment are also presented, and the prospects for the development and improvement of the system are discussed. The software is implemented using the ROS2 software platform.