EVOLUTION OF SEMANTIC SEGMENTATION IN VISUAL SLAM: INTEGRATION OF SINGLE-STAGE DETECTORS AND ADAPTIVE FEATURE WEIGHTING METHOD

Abstract

This paper examines the
evolution of semantic segmentation methodologies in simultaneous localization and mapping (SLAM)
for dynamic environments. The paradigm shift from computationally intensive two-stage architectures
(Mask R-CNN) to streamlined single-stage models (YOLO), which facilitate real-time edge computing,
is analyzed. A fundamental deficiency in extant systems is identified: the application of rigid (binary)
feature rejection, causing the deleterious omission of geometric data from spatially static entities within
dynamic taxonomies. To ameliorate this limitation, the Soft-Penalty Model is postulated. This
mathematical construct executes adaptive modifications upon the noise covariance matrix in the Kalman
filter, operationalized via the neural network's semantic confidence index
Keywords: SLAM, semantic segmentation, Kalman filter, dynamic environment, localization,
unmanned systems, convolutional neural networks.

References
1. Mur-Artal R., Montiel J. M. M., Tardos J. D. ORB-SLAM: A Versatile and Accurate
Monocular SLAM System. IEEE Transactions on Robotics. 2015. Vol. 31, no. 5. P. 1147–1163.
URL: https://doi.org/10.1109/tro.2015.2463671.
2. Choi K.-S., Lee S.-G. Enhanced SLAM for a mobile robot using extended Kalman Filter
and neural networks. International Journal of Precision Engineering and Manufacturing. 2010. Vol.
11, no. 2. P. 255–264. URL: https://doi.org/10.1007/s12541-010-0029-9.
3. Bescos B., Fácil J. M., Civera J., Neira J. DynaSLAM: Tracking, Mapping, and Inpainting
in Dynamic Scenes. IEEE Robotics and Automation Letters. 2018. Vol. 3, no. 4. P. 4076–4083.
4. Wadud, R. A., & Sun, W. (2022). DyOb-SLAM: Dynamic Object Tracking SLAM System.
arXiv. URL: https://doi.org/10.48550/arXiv.2211.01941.
5. Jocher G. et al. YOLO by Ultralytics (Version 11). 2024. URL:
https://github.com/ultralytics/ultralytics.
6. Wang C. et al. Applying SLAM Algorithm Based on Nonlinear Optimized Monocular
Vision and IMU in the Positioning Method of Power Inspection Robot in Complex Environment.
Mathematical Problems in Engineering. 2022. Vol. 2022. P. 1–14. URL:
https://doi.org/10.1155/2022/3378163Kim, A., Osep, A., & Leal-Taixe, L. (2021). EagerMOT: 3D
Multi-Object Tracking via Sensor Fusion.
7. Liu H. Identifying and updating local optimization methods in extended Kalman filter
SLAM. Applied and Computational Engineering. 2023. Vol. 4, no. 1. P. 569–573. URL:
https://doi.org/10.54254/2755-2721/4/2023325