TRAVEL: Traversable Ground and Above-Ground Object Segmentation Using Graph Representation of 3D LiDAR Scans

Abstract: Perception of traversable regions and objects of interest from a 3D point cloud is one of the critical tasks in autonomous navigation. A ground vehicle needs to look for traversable terrains that are explorable by wheels. Then, to make safe navigation decisions, the segmentation of objects positioned on those terrains has to be followed up. However, over-segmentation and under-segmentation can negatively influence such navigation decisions. To that end, we propose TRAVEL, which performs traversable ground detection and object clustering simultaneously using the graph representation of a 3D point cloud. To segment the traversable ground, a point cloud is encoded into a graph structure, tri-grid field, which treats each tri-grid as a node. Then, the traversable regions are searched and redefined by examining local convexity and concavity of edges that connect nodes. On the other hand, our above-ground object segmentation employs a graph structure by representing a group of horizontally neighboring 3D points in a spherical-projection space as a node and vertical/horizontal relationship between nodes as an edge. Fully leveraging the node-edge structure, the above-ground segmentation ensures real-time operation and mitigates over-segmentation. Through experiments using simulations, urban scenes, and our own datasets, we have demonstrated that our proposed traversable ground segmentation algorithm outperforms other state-of-the-art methods in terms of the conventional metrics and that our newly proposed evaluation metrics are meaningful for assessing the above-ground segmentation. We will make the code and our own dataset available to public at https://github.com/url-kaist/TRAVEL.

• The paper introduces a novel graph-based segmentation algorithm using a Tri-Grid Field for precise delineation of traversable ground.
• It employs spherical projection with breadth-first search to segment above-ground objects and mitigate over-segmentation.
• The method outperforms traditional approaches on datasets like CARLA and Semantic KITTI, enhancing autonomous navigation performance.

An Analytical Overview of "TRAVEL: Traversable Ground and Above-Ground Object Segmentation Using Graph Representation of 3D LiDAR Scans"

The paper "TRAVEL: Traversable Ground and Above-Ground Object Segmentation Using Graph Representation of 3D LiDAR Scans" introduces TRAVEL, a novel segmentation algorithm aimed at enhancing the perception capabilities essential for autonomous navigation. The authors target the critical task of segmenting traversable ground and above-ground objects within a 3D LiDAR-generated point cloud. This research outlines a dual-phase approach using graph representations to improve segmentation accuracy and computational efficiency.

Key Contributions

The primary contribution of the paper is its node-edge graph representation of LiDAR data, known as the Tri-Grid Field (TGF), which facilitates effective ground segmentation. This graph-based model allows the TRAVEL algorithm to efficiently discern traversable terrains by evaluating local convexity and concavity between connected nodes. The research posits that encoded graph representations enhance the segmentation accuracy by ensuring a nuanced understanding of the geometric relationships of terrain elements.

Another significant contribution is the above-ground object segmentation technique, which employs spherical projection. This representation treats horizontally neighboring segments of 3D points as nodes and establishes vertical and horizontal connections between these nodes. The algorithm adapts a breadth-first search within this framework to mitigate over-segmentation.

Methodological Insights

TRAVEL's methodology emphasizes robustness to varying environments, a key requirement for autonomous navigation systems. Its TGF-based segmentation distinguishes it from conventional plane-based algorithms that often fail to address traversability. The proposed algorithm is demonstrated to outperform traditional methods, showing improved results on datasets like CARLA and Semantic KITTI.

In the empirical evaluation, TRAVEL's performance is quantified using both standard metrics (precision, recall, F1-score, and accuracy) and novel metrics (over-segmentation entropy and under-segmentation entropy) designed by the researchers to better assess segmentation quality. The use of entropy-based metrics reflects a deeper analysis of segmentation performance, accounting for distribution and uncertainty within object labels.

Implications and Future Directions

The implications of the TRAVEL algorithm are broad, particularly in domains where autonomous systems are expected to operate in diverse environments. The approach suggests that graph-based segmentation can significantly enhance the perception layers in robotics and autonomous vehicles by accurately delineating navigable terrains and identifying potential obstacles.

Looking forward, the authors suggest future work might incorporate TRAVEL into navigation tasks that involve dynamic object identification and removal. Integration with learning-based models could also be explored to balance real-time performance with class identification, potentially extending TRAVEL's applicability to environments with more complex object interactions.

In summary, the paper showcases TRAVEL as a significant advance in segmentation methodologies, driven by its graph representation of LiDAR scans and its balanced focus on computational efficiency and segmentation accuracy. Its approach provides a solid foundation for ongoing research and practical applications in autonomous navigation technologies.