.png){kind=icon}
.png){kind=icon}
Return
ORB-SLAM3 with dogleg-based improved graph optimization
.png){kind=icon}
PRE
AI
DOI：10.1088/2631-8695/ae2386.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
To solve the problems of easy divergence and premature convergence of ORB-SLAM3 in the process of localization and map building, this paper introduces the dogleg algorithm to replace the traditional LM optimization method and improves the back-end nonlinear optimization module. On the basis of keeping the overall framework of the system unchanged, the algorithm is carefully adjusted and debugged in terms of the parameter settings of the dogleg optimizer and the initialization of the Hessian matrix. The algorithm first collects data using the tracking thread module, then constructs the local graph and performs local beam adjustment and dogleg optimization and finally performs historical key-frames matching and loop identification using the loopback detection module. Experimental results show that the improved system outperforms the original system in terms of objective indices such as ATE and RPE, especially in the repositioning and closed-loop optimization phases, which verifies the application value and practical effect of the dogleg algorithm in visual SLAM back-end optimization.
Keywords:
SLAM
visual SLAM
dogleg method
Journal
E
IF:
1.6
Papers:
2.1K
Citations:
0