Crowd-Sourced Mapping of New Feature Layer for High-Definition Map

Abstract

A High-Definition map (HD map) is a precise and detailed map composed of various landmark feature layers. The HD map is a core technology that facilitates the essential functions of intelligent vehicles. Recently, it has come to be required for the HD map to continuously add new feature layers in order to increase the performances of intelligent vehicles in more complicated environments. However, it is difficult to generate a new feature layer for the HD map, because the conventional method of generating the HD map based on several professional mapping cars has high costs in terms of time and money due to the need to re-drive on all of the public roads. In order to reduce these costs, we propose a crowd-sourced mapping process of the new feature layer for the HD map. This process is composed of two steps. First, new features in the environments are acquired from multiple intelligent vehicles. The acquired new features build each new feature layer in each intelligent vehicle using the HD map-based GraphSLAM approach, and these new feature layers are conveyed to a map cloud through a mobile network system. Next, the crowd-sourced new feature layers are integrated into a new feature layer in a map cloud. In the simulation, the performance of the crowd-sourced process is then analyzed and evaluated. Experiments in real driving environments confirm the results of the simulation.

Mobile Edge Intelligence and Computing for the Internet of Vehicles

“…It is expected that HD map-based localization will be a common approach in Level 4 and Level 5 autonomous driving systems. HD map also helps to resolve two main issues of perception based on on-board sensors [59]:…”

A Geodetic Normal Distribution Map for Long-Term LiDAR Localization on Earth

“…Among them, map features for map-matching localization can be split by sensor type: camera and LiDAR. For camera-based localization [16]- [19], the features detected by the camera can be matched with the semantic features of the HD map, such as lanes [5], road surface markings [6], and traffic control devices [7]. Camera-based localization has the following advantages.…”

“…Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2020.3047421, IEEE Access tures measured by the camera can be matched with the semantic features of the HD map, such as lanes [5], road surface markings [6], and traffic control devices [7]. However, it is known that camera-based localization does not satisfy the localization performance in some urban environments with few semantic features such as alley and intersection.…”

Lane-Level Road Network Generation Techniques for Lane-Level Maps of Autonomous Vehicles: A Survey

“…In this approach, the sample frequency of the GPS is 5−120 s, and the average sample frequency is 5−60 s. Accordingly, the accuracy for samples is meters to tens of meters. For example, U-blox EVK-6T is low cost, and the position accuracy is 2.5 m [30]. In addition, the Inertial Navigation System (INS) provides the positions of vehicles with Inertial Measurement Units (IMU) with gyro and accelerometers.…”