基于图优化的紧耦合双目视觉/惯性/激光雷达SLAM方法

Abstract

基于单一传感器的同时定位与地图构建技术已经逐渐不能满足移动机器人、无人机及自动驾驶车辆等智能移动载体日益复杂的应用场景。为了进一步提升移动载体在复杂环境下的定位与建图性能，基于多传感器融合的SLAM技术成为目前研究的热点内容。本文提出了一种基于图优化的紧耦合双目视觉/惯性/激光雷达SLAM方法(S-VIL SLAM)，该方法在视觉惯性系统中引入激光雷达原始观测，基于滑动窗口实现了IMU量测、视觉特征及激光点云特征的多源数据联合非线性优化。利用视觉与激光雷达的互补特性设计了视觉增强的激光雷达闭环优化算法，进一步提升了多源融合SLAM系统的全局定位与建图精度。为了验证本文算法的性能，利用自主搭建的集成多传感器的硬件采集平台在室外场景下进行了车载试验。试验结果表明，本文提出的紧耦合双目视觉/惯性/激光雷达里程计相比于紧耦合双目视觉惯性里程计和激光雷达里程计定位定姿性能显著提升，视觉增强的激光雷达闭环优化算法能够在大尺度场景下有效探测出轨迹中的闭环信息，并实现高精度的全局位姿图优化，经过闭环优化的点云地图具有良好的分辨率和全局一致性。

Alternate abstract:

Simultaneous localization and mapping (SLAM) technology based on a single sensor has gradually been unable to meet the increasingly complex application scenarios of the intelligent mobile carriers such as mobile robots,unmanned aerial vehicles,and self-driving cars.In order to further improve the localization and mapping performance of the mobile carriers in complex environments,multi-sensor fusion SLAM has become a hotspot of current research.In this contribution,we present a graph-optimization based and tightly-coupled stereo visual-inertial-LiDAR SLAM termed S-VIL SLAM,which integrates the LiDAR observations into a visual-inertial system.In this work,the IMU measurements,visual features,and laser point cloud features are jointly optimized in a sliding window.Moreover,a vision enhanced loop-closure algorithm of LiDAR is designed in this paper by using the complementary characteristics between vision and LiDAR,which further improves the global positioning and mapping accuracy of the multi-sensor fusion SLAM.We perform vehicle-borne experiments in outdoor environments to assess the performance of the proposed approach.The experimental results indicate that the proposed S-VIL odometry outperforms the state-of-the-art tightly coupled visual-inertial odometry (VIO) and LiDAR odometry in terms of pose estimation accuracy.The proposed loop-closure algorithm can effectively detect the loop closure of trajectories in large-scale scenes and achieve high-precision pose graph optimization.The point cloud map after loop closure optimization has good resolution and global consistency.