TOSS: Real-time Tracking and Moving Object Segmentation for Static Scene Mapping

TL;DR

The paper addresses the need for real-time autonomous navigation by jointly handling moving-object segmentation (MOS) and static map building in dynamic environments. It introduces TOSS, a real-time MOS framework that fuses online object tracking with static-map construction, featuring a hierarchical association cost matrix that reduces data-association complexity from $O(N^2)$ to $O(kN)$ and a DS-Voting refinement that leverages spatio-temporal cues to improve dynamic/static labeling. The approach is validated on SemanticKITTI and challenging real-world datasets, showing superior Preservation Rate (PR), competitive Rejection Rate (RR), and robust performance under pose inaccuracies, with real-time operation demonstrated via reduced runtimes compared to exhaustive methods. Overall, TOSS enables robust, real-time MOS and static map creation in unstructured environments, supporting safer navigation and higher-quality maps for legged-robot platforms.

Abstract

Safe navigation with simultaneous localization and mapping (SLAM) for autonomous robots is crucial in challenging environments. To achieve this goal, detecting moving objects in the surroundings and building a static map are essential. However, existing moving object segmentation methods have been developed separately for each field, making it challenging to perform real-time navigation and precise static map building simultaneously. In this paper, we propose an integrated real-time framework that combines online tracking-based moving object segmentation with static map building. For safe navigation, we introduce a computationally efficient hierarchical association cost matrix to enable real-time moving object segmentation. In the context of precise static mapping, we present a voting-based method, DS-Voting, designed to achieve accurate dynamic object removal and static object recovery by emphasizing their spatio-temporal differences. We evaluate our proposed method quantitatively and qualitatively in the SemanticKITTI dataset and real-world challenging environments. The results demonstrate that dynamic objects can be clearly distinguished and incorporated into static map construction, even in stairs, steep hills, and dense vegetation.

Figures (6)

• Figure 1: Overview of TOSS. TOSS is a multi-step process that involves (a) segmenting ground and instance points (see Sec. \ref{['sec:travel']}), (b) real-time object tracking to classify them as coarse dynamic or static objects (see Sec. \ref{['sec:mot']}), and finally (c) a voting-based refinement module for accurate dynamic object removal and static object recovery on the static map (see Sec. \ref{['sec:dsvote']}).
• Figure 2: Results comparison for dynamic and static objects.The first and third rows of the figure represent the results without using DS-Voting, while the second and fourth rows depict the results after using DS-Voting. false positive and false negative points are corrected by DS-Voting.
• Figure 3: Our test environment for challenging environments in KAIST campus, Republic of Korea. Left test site ranges from flat road to stairways and steep hills in the wild, while right one contains various bumpy terrains.
• Figure 4: Our quadrupedal platforms, Go1 and A1 from Unitree Robotics, are equipped with one range sensor (Ouster OS0-128) and one IMU sensor (Xsens MTI-300).
• Figure 5: Mapping result comparison SemanticKITTI sequence 02, 07. Qualitative static map results ERASOR, OctoMap, and Ours. ERASOR is map cleaning method and OctoMap and Ours are map update method. Green points indicate true positive (TP), red points indicate false positive (FP), and blue indicates false negative (FN).