The SLAM system based on static scene assumption will introduce huge estimation errors when moving objects appear in the field of view. This paper proposes a novel multi-object dynamic lidar odometry (MLO) based on semantic object detection technology to solve this problem. The MLO system can provide reliable localization of robot and semantic objects and build long-term static maps in complex dynamic scenes. For ego-motion estimation, we use the environment features that take semantic and geometric consistency constraints into account in the extraction process. The filtering features are robust to semantic movable and unknown dynamic objects. At the same time, a least square estimator using the semantic bounding box and object point cloud is proposed to achieve accurate and stable multi-object tracking between frames. In the mapping module, we further realize dynamic semantic object detection based on the absolute trajectory tracking list (ATTL). Then, static semantic objects and environmental features can be used to eliminate accumulated localization errors and build pure static maps. Experiments on public KITTI data sets show that the proposed system can achieve more accurate and robust tracking of the object and better real-time localization accuracy in complex scenes compared with existing technologies.

翻译：基于静态场景假设的SLAM系统在视野中出现运动物体时会引入巨大的估计误差。本文提出一种基于语义目标检测技术的多目标动态激光雷达里程计（MLO）来解决这一问题。MLO系统能够在复杂动态场景中为机器人及语义目标提供可靠的定位，并构建长期静态地图。在自运动估计中，我们采用提取过程中兼顾语义与几何一致性约束的环境特征。滤波后的特征对语义可移动及未知动态物体具有鲁棒性。同时，提出一种基于语义边界框和目标点云的最小二乘估计器，以实现帧间精确稳定的多目标跟踪。在建图模块中，我们进一步基于绝对轨迹跟踪列表（ATTL）实现动态语义目标检测。随后利用静态语义目标和环境特征消除累积定位误差，并构建纯静态地图。在公开KITTI数据集上的实验表明，与现有技术相比，所提系统能够在复杂场景中实现更精确稳健的目标跟踪和更优的实时定位精度。