In dynamic scenes, both localization and mapping in visual SLAM face significant challenges. In recent years, numerous outstanding research works have proposed effective solutions for the localization problem. However, there has been a scarcity of excellent works focusing on constructing long-term consistent maps in dynamic scenes, which severely hampers map applications. To address this issue, we have designed a multi-level map construction system tailored for dynamic scenes. In this system, we employ multi-object tracking algorithms, DBSCAN clustering algorithm, and depth information to rectify the results of object detection, accurately extract static point clouds, and construct dense point cloud maps and octree maps. We propose a plane map construction algorithm specialized for dynamic scenes, involving the extraction, filtering, data association, and fusion optimization of planes in dynamic environments, thus creating a plane map. Additionally, we introduce an object map construction algorithm targeted at dynamic scenes, which includes object parameterization, data association, and update optimization. Extensive experiments on public datasets and real-world scenarios validate the accuracy of the multi-level maps constructed in this study and the robustness of the proposed algorithms. Furthermore, we demonstrate the practical application prospects of our algorithms by utilizing the constructed object maps for dynamic object tracking.

翻译：在动态场景中，视觉SLAM的定位与建图面临显著挑战。近年来，大量优秀研究工作已针对定位问题提出了有效解决方案，但针对动态场景下构建长期一致地图的杰出工作仍较为匮乏，这严重制约了地图的实际应用。为解决这一问题，我们设计了一套面向动态场景的多层级地图构建系统。在该系统中，我们采用多目标跟踪算法、DBSCAN聚类算法及深度信息修正目标检测结果，精准提取静态点云，并构建稠密点云地图与八叉树地图。针对动态场景，我们提出了一种平面地图构建算法，涵盖动态环境中平面的提取、滤波、数据关联及融合优化，从而生成平面地图。此外，我们引入了一种面向动态场景的物体地图构建算法，包括物体参数化、数据关联及更新优化。在公开数据集与真实场景中的大量实验验证了本研究构建的多层级地图的准确性及所提算法的鲁棒性。进一步地，我们通过利用构建的物体地图进行动态目标跟踪，展示了所提算法的实际应用前景。