Semantic Mapping Based on Visual SLAM with Object Model Replacement Visualization for Cleaning Robot

Cleaning robots have been widely used in our daily life, achieving huge success in the service robot sector. The autonomous movement and decision-making of these robots directly depend on their environmental perception capability, which is implemented using SLAM (simultaneous localization and mapping) technique. In order to achieve better environmental perception and intelligent decision-making, it is necessary to introduce visual SLAM into cleaning robot and take full advantage of visual information to achieve higher-level perception, instead of sticking to current feature point or pixel level perception capability. In this paper, we propose a framework for semantic mapping with object model replacement visualization in indoor scenes for cleaning robot. We use an EKF-based method to fuse wheel odometry and visual odometry, and build a map representation for navigation in the framework. We obtain a semantic map by fusing object detection and point cloud segmentation. We also develop object model replacement for better visualization purposes. Finally, we conduct dataset tests and real experiments to validate our proposed framework.