基于语义栅格地图的长走廊退化环境SLAM

Semantic Grid Map-Based SLAM for Degraded Long-Corridor Environments

摘要:

针对长走廊退化环境中纹理稀疏、光照不足以及环境特征弱等问题，对一种基于语义栅格地图的 SLAM 方法进行了研究，通过引入语义信息强化环境表达能力，并利用多传感器融合提升位姿估计精度，以突破退化环境中定位不稳定与地图表达不足的局限。采用 EKF 融合 IMU、轮式里程计和激光里程计数据以获得更高精度的位姿估计，结合目标检测技术提取物体类别与空间位置等语义信息，并与栅格地图整合生成语义栅格地图。经对比实验分析，该方法在地图精度、环境结构一致性以及语义表达完整性方面均优于传统栅格地图，有效抑制了退化环境中因特征不足导致的重影与漂移现象。同时，融合语义信息的地图能够清晰呈现环境中关键目标的语义类别与位置信息，为后续导航、检索或任务执行提供更具价值的环境支持。

Abstract:

To address issues such as sparse textures, insufficient illumination, and weak environmental features in degraded long-corridor environments, a SLAM method based on semantic grid mapping was investigated, in which semantic information was incorporated to enhance environmental representation and multi-sensor fusion was employed to improve pose estimation accuracy, thereby overcoming limitations of unstable localization and insufficient map representation in degraded scenarios. An Extended Kalman Filter (EKF) was used to fuse IMU, wheel odometry, and LiDAR odometry data to obtain high-precision pose estimation, while object detection techniques extracted semantic information such as object categories and spatial locations, which was integrated with the grid map to generate a semantic grid map. Comparative experimental analysis shows that the method outperforms traditional grid mapping in mapping accuracy, structural consistency, and semantic completeness, effectively suppressing ghosting and drift caused by insufficient features in degraded environments. In addition, the semantic-enhanced map clearly presents the semantic categories and spatial information of key objects in the environment, providing enhanced environmental information support for subsequent navigation, retrieval, and task execution.