Robust Ego-motion Estimation and Map Matching Technique for Autonomous Vehicle Localization with High Definition Digital Map

Abstract

One of the essential technologies required for environmental recognition of an autonomous vehicle is a localization technique that recognizes the position and orientation of the vehicle. In contrast to previous localization techniques that generate map data from sensor data itself, there is an increasing number of studies using high definition (HD) digital maps. The map-based localization technology consists of predicting the position of the next step through the ego-motion of the vehicle and determining the position through map matching. In this paper, we propose a robust ego-motion estimation and map matching technology for robust vehicle localization. First, we propose a visual odometry (VO) model for robust ego-motion estimation and a vehicle planar motion model based on in-vehicle sensors to improve the robustness of VO in the absence of image features. We also introduce a new line segmentation matching model and a geometric correction method of extracted road marking from an inverse perspective mapping (IPM) for robust map matching techniques. The technology proposed in this paper has been verified in various ways through real autonomous vehicles and successfully acquired the autonomous driving license of the Republic of Korea.