Road Curb Detection: ADAS for a Road Sweeper Vehicle

Abstract

Recent development of the concept of smart cities has led to an increasing demand for advanced technological solutions that drive forward the design and capabilities of utility vehicles operating in urban environments. One possibility is to exploit recent advances in computer vision to introduce a certain level of autonomy into some of the vehicle’s functionalities, e.g., an advanced driver-assistance system. Modern road sweeper vehicles are designed to possess multiple systems for maintaining the road quality and city cleanliness, such as brushes, vacuums, and great vehicle maneuverability. Introducing autonomy to these control systems lowers the burden on the human operator, thereby increasing work efficiency and overall safety, as well as making a positive impact on worker health. This paper considers a 3D curb detection system that supports autonomous road sweeping. In order to achieve this, we utilize a vision-based approach that leverages stereo depth estimation and a pre-trained semantic segmentation model. In addition, we implement a simple LiDAR-based curb detection baseline. Finally, we collected our own dataset comprised of driving sequences resembling our use-case, which is used to conduct qualitative experiments.