.png){kind=icon}
.png){kind=icon}
Return
Optimization RGB-D 3-D Reconstruction Algorithm Based on Dynamic SLAM
.png){kind=icon}
PRE
AI
DOI：10.1109/TIM.2023.3248116.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This article proposes a robust dynamic feature segmentation simultaneous localization and mapping (SLAM) 3-D reconstruction algorithm based on the dynamic segmentation framework of DynaSLAM in order to realize the SLAM system mapping in the dynamic environment of intelligent robots. The algorithm employs the method of feature point reextraction to solve the problem of an insufficient number of original feature points. In the preprocessing stage, the optimized epipolar geometry model is designed in this article, and the image is segmented jointly with Mask R-CNN. At the point cloud postprocessing stage, a kernel principal component analysis (PCA) is used to reduce noise, and then, this article designs a dynamic filtering method based on octree for further processing. This processing effectively eliminates additional dynamic outliers. The results of experiments on TUM and Bonn datasets show that this algorithm achieves more than 10% better attitude estimation accuracy than the DynaSLAM algorithm in a highly dynamic environment and has advantages over the state-of-the-art (SOTA) algorithms.
Keywords:
Simultaneous localization and mapping
Feature extraction
Heuristic algorithms
Image segmentation
Geometry
Cameras
Location awareness
3-D reconstruction
dynamic segmentation
red green blue-depth (RGB-D)
simultaneous localization and mapping (SLAM)
Journal
IF:
5.9
Papers:
1.9W
Citations:
5.8W