ORB-SLAM2 System Enhanced With SiaT-Hough Module on Mars Digital Twin Platform: For Rock Counting and 3D Reconstruction in Martian Environments

ORB-SLAM2 System Enhanced With SiaT-Hough Module on Mars Digital Twin Platform: For Rock Counting and 3D Reconstruction in Martian Environments

This paper proposes a digital twin platform for Mars environment based on NVIDIA Isaac Sim to create a unique random environment with multi-scale fine-grained terrain features through programmed terrain generation and asset management. The SiaT-Hough module was added to the ORB-SLAM2 system for Mars...

Full description

Saved in:
Bibliographic Details
Main Authors: Wei Liu, Gang Wan, Jia Liu, Dianwei Cong
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Mars exploration
digital twin
ORB-SLAM2
semantic segmentation
Online Access:https://ieeexplore.ieee.org/document/11071669/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This paper proposes a digital twin platform for Mars environment based on NVIDIA Isaac Sim to create a unique random environment with multi-scale fine-grained terrain features through programmed terrain generation and asset management. The SiaT-Hough module was added to the ORB-SLAM2 system for Mars rock segmentation and 3D modeling, as well as rock counting. We propose a lightweight Siamese Transformer (SiaT) network combined with Hough transform for small sample semantic segmentation tasks, especially for rock counting in Mars exploration scenarios. Experimental results demonstrate that the SiaT-Hough model achieves an average segmentation accuracy of 87.8<sc>6</sc>% and a processing speed of 57.34 FPS on the Mars simulation dataset, striking an optimal balance between accuracy and real-time performance. In addition, we also carried out 3D reconstruction positioning accuracy evaluation and point cloud quality evaluation, and the positioning absolute track error on the two public data sets was less than 0.1m, 0.0639m and 0.0106m, respectively. The Pearson linear correlation coefficient of mean point cloud mass in Mars rendering dataset reaches 0.852, and the mean relative error of rock count is 8.63%, which proves the feasibility of the improved ORB-SLAM2 system for rock count in Mars environment.
ISSN:2169-3536

Similar Items