Towards accurate instance segmentation in large-scale LiDAR point clouds

TL;DR

This paper improves large-scale LiDAR point cloud instance segmentation by developing a clustering strategy that leverages multiple learned embeddings, significantly enhancing the separation of nearby objects in outdoor scene understanding tasks.

Contribution

It introduces a novel clustering approach using multiple learned embeddings to improve instance segmentation accuracy in large-scale LiDAR point clouds.

Findings

Significant improvement in separating adjacent instances.

Effective on urban and forest datasets.

Versatile clustering strategy enhances panoptic segmentation.

Abstract

Panoptic segmentation is the combination of semantic and instance segmentation: assign the points in a 3D point cloud to semantic categories and partition them into distinct object instances. It has many obvious applications for outdoor scene understanding, from city mapping to forest management. Existing methods struggle to segment nearby instances of the same semantic category, like adjacent pieces of street furniture or neighbouring trees, which limits their usability for inventory- or management-type applications that rely on object instances. This study explores the steps of the panoptic segmentation pipeline concerned with clustering points into object instances, with the goal to alleviate that bottleneck. We find that a carefully designed clustering strategy, which leverages multiple types of learned point embeddings, significantly improves instance segmentation. Experiments on the NPM3D urban mobile mapping dataset and the FOR-instance forest dataset demonstrate the effectiveness and versatility of the proposed strategy.