基于强度扫描上下文回环检测的激光雷达SLAM算法

Lidar SLAM based on intensity scan context loop closure detection

基于激光雷达的同步定位与建图(SLAM)技术可以在未知环境中实现机器人的实时定位并构建环境地图。LOAM、Le GO-LOAM等经典激光SLAM算法仅依赖点云的几何信息进行位姿估计,忽略了强度信息具有有效位置识别的特性;同时采用迭代计算的方式进行点云畸变补偿,虽然精度有保证但带来了高消耗的计算。基于此,提出了一种基于强度扫描上下文回环检测的激光SLAM算法,同时利用点云的几何和强度信息,采用强度扫描上下文(ISC)作为全局描述符进行回环检测以减少漂移误差,此外,采用非迭代两步法实现点云畸变补偿,以降低计算成本。基于室外公开数据集和室内采集数据的实验表明,所提出的激光雷达SLAM算法可有效抑制里程计位姿漂移,相比仅利用点云几何信息位姿精度平均提高约50%(均方根值),并在增加回环检测模块的情况下保证算法的实时性。

Lidar simultaneous localization and mapping(SLAM) can realize real-time positioning of robots and build environmental maps in unknown environments, and has received extensive attention in recent years.Classic laser SLAM algorithms such as LOAM and Lego-LOAM only rely on the geometric information of the point cloud for pose estimation, ignoring the uniqueness of the intensity information and can also be used for effective position recognition. In addition, the iterative calculation methods in these classical algorithms are used to compensate point cloud distortion. Although the accuracy is guaranteed, the high consumption calculation is cost. Therefore, a lidar SLAM based on the intensity scan context loop closure detection is proposed. At the same time, the geometric and intensity information of the point cloud are utilized, and intensity scan context(ISC) is used as the global descriptor for loop closure detection to reduce the drift error.In addition, point cloud distortion compensation is implemented using a non-iterative two-step method to reduce computational cost. Experiments based on outdoor open data sets and indoor data collection show that the proposed laser SLAM algorithm can effectively suppress the odometry pose drift of odometer, improve the pose accuracy by about 50% on average compared with only using point cloud geometric information,which ensure the real-time performance of the algorithm when the loop closure detection module is added.