Journal Article

PLMOT-SLAM: a point-line features fusion SLAM system with moving object tracking

Yao, Xifan, Wang, Kesai, Ran, Guangjun, Ma, Nanfeng

2025

Overview

In this study, we propose a simultaneous localization and mapping (SLAM) system that integrates point-line features fusion with moving object tracking to improve pose estimation accuracy and semantic representation capabilities under challenging conditions. Utilizing the real-time tracking count of feature points and the average length of line segments, we develop an adaptive line feature selection algorithm and a dynamic point-line feature joint optimization function to enhance the efficiency of point-line fusion. Furthermore, we incorporate a motion check module based on spatial geometric constraints with the object tracking algorithm called DeepSort, to enhance the robustness of dynamic object segmentation. Through experimental analysis, we validate the significance of line feature quality in point-line fusion, as well as the necessity of moving object tracking for improving the accuracy of pose estimation in dynamic environments. Comprehensive experiments conducted using the public indoor datasets show significant improvements over the original system. Moreover, compared to other state-of-the-art SLAM systems, our approach demonstrates advantages in both pose estimation and system functionality. The datasets and the source code are available at https://github.com/kesai0518/Adaptive-Point-Line-Fusion-SLAM-with-MOT .

• Publisher: Springer Berlin Heidelberg,Springer Nature B.V
• Subject: Accuracy / Algorithms / Artificial Intelligence / Cameras / Computer Graphics / Computer Science / Datasets / Deep learning / Feature selection / Geometric constraints / Image Processing and Computer Vision / Localization / Methods / Moving object recognition / Optimization / Pose estimation / Real time / Semantics / Simultaneous localization and mapping / Source code / Tracking / Unmanned aerial vehicles