A Visual SLAM Algorithm for Indoor Dynamic Scenes Based on Semantic Feature Screening

Yao Wang, Changzhong Pan, Hao Huang
Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, PMLR 278:155-163, 2025.

Abstract

This study innovatively proposes a dynamic scene SLAM al- gorithm that integrates semantic feature filtering mechanism to address the problems of large positioning deviation and dense map artifacts in visual SLAM systems in indoor dynamic environments. In the track- ing module of the ORB-SLAM3 framework, this algorithm introduces a lightweight YOLOv11-seg neural network for scene semantic analysis and target area calibration. Through semantic information, depth data, and geometric relationships, feature point motion state discrimination is achieved, and a high-precision dynamic feature filtering algorithm is developed. To verify the performance of the algorithm, benchmark tests were conducted on the TUM dataset of the Technical University of Mu- nich. Comparative experimental data showed that in the high dynamic conditions of the TUM testing scenario, this scheme achieved significant improvement in trajectory tracking accuracy, and its positioning per- formance was significantly better than the current mainstream dynamic SLAM technology.

Cite this Paper

BibTeX
@InProceedings{pmlr-v278-wang25b, title = {A Visual SLAM Algorithm for Indoor Dynamic Scenes Based on Semantic Feature Screening}, author = {Wang, Yao and Pan, Changzhong and Huang, Hao}, booktitle = {Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing}, pages = {155--163}, year = {2025}, editor = {Zeng, Nianyin and Pachori, Ram Bilas and Wang, Dongshu}, volume = {278}, series = {Proceedings of Machine Learning Research}, month = {25--27 Apr}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v278/main/assets/wang25b/wang25b.pdf}, url = {https://proceedings.mlr.press/v278/wang25b.html}, abstract = {This study innovatively proposes a dynamic scene SLAM al- gorithm that integrates semantic feature filtering mechanism to address the problems of large positioning deviation and dense map artifacts in visual SLAM systems in indoor dynamic environments. In the track- ing module of the ORB-SLAM3 framework, this algorithm introduces a lightweight YOLOv11-seg neural network for scene semantic analysis and target area calibration. Through semantic information, depth data, and geometric relationships, feature point motion state discrimination is achieved, and a high-precision dynamic feature filtering algorithm is developed. To verify the performance of the algorithm, benchmark tests were conducted on the TUM dataset of the Technical University of Mu- nich. Comparative experimental data showed that in the high dynamic conditions of the TUM testing scenario, this scheme achieved significant improvement in trajectory tracking accuracy, and its positioning per- formance was significantly better than the current mainstream dynamic SLAM technology.} }
Endnote
%0 Conference Paper %T A Visual SLAM Algorithm for Indoor Dynamic Scenes Based on Semantic Feature Screening %A Yao Wang %A Changzhong Pan %A Hao Huang %B Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing %C Proceedings of Machine Learning Research %D 2025 %E Nianyin Zeng %E Ram Bilas Pachori %E Dongshu Wang %F pmlr-v278-wang25b %I PMLR %P 155--163 %U https://proceedings.mlr.press/v278/wang25b.html %V 278 %X This study innovatively proposes a dynamic scene SLAM al- gorithm that integrates semantic feature filtering mechanism to address the problems of large positioning deviation and dense map artifacts in visual SLAM systems in indoor dynamic environments. In the track- ing module of the ORB-SLAM3 framework, this algorithm introduces a lightweight YOLOv11-seg neural network for scene semantic analysis and target area calibration. Through semantic information, depth data, and geometric relationships, feature point motion state discrimination is achieved, and a high-precision dynamic feature filtering algorithm is developed. To verify the performance of the algorithm, benchmark tests were conducted on the TUM dataset of the Technical University of Mu- nich. Comparative experimental data showed that in the high dynamic conditions of the TUM testing scenario, this scheme achieved significant improvement in trajectory tracking accuracy, and its positioning per- formance was significantly better than the current mainstream dynamic SLAM technology.
APA
Wang, Y., Pan, C. & Huang, H.. (2025). A Visual SLAM Algorithm for Indoor Dynamic Scenes Based on Semantic Feature Screening. Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, in Proceedings of Machine Learning Research 278:155-163 Available from https://proceedings.mlr.press/v278/wang25b.html.

Related Material