Semantic-Enhanced LiDAR-Inertial Odometry for Autonomous Driving in Urban Environments

Odometry estimation is a fundamental technique for many robotics and autonomous driving applications. Recent advances using LiDAR and IMU sensor-fusion methods show promising potential in odometry estimation by combining the precise perception of LiDAR and the high-frequency motion estimation from IMU sensors. In this paper, we propose a novel semantic-enhanced LiDAR-inertial odometry (LIO) algorithm for autonomous vehicles working in urban driving environments. It extracts specific semantics needed by odometry, the static flat features and ground constraints, to improve the accuracy and the robustness of LIO. To this end, our method exploits a deep neural network to detect the static points in the LiDAR scan and based on that extracts flat features. It furthermore uses ground plane information to eliminate the z-axis drifts in the odometry estimates. We thoroughly evaluated our method using multiple publicly available datasets and also a self-recorded dataset. We use the network trained only on one dataset and directly apply our method to other datasets without fine-tuning. The experimental results show that our method is accurate and reliable in dynamic urban environments and achieves state-of-the-art LIO performance with fast speed and strong generalization ability.