Incremental and Enhanced Scanline-Based Segmentation Method for Surface Reconstruction of Sparse LiDAR Data

Weimin Wang,Nobuo Kawaguchi,Ken Sakurada

doi:10.3390/rs8110967

Abstract

The segmentation of point clouds is an important aspect of automated processing tasks such as semantic extraction. However, the sparsity and non-uniformity of the point clouds gathered by the popular 3D mobile LiDAR devices pose many challenges for existing segmentation methods. To improve the segmentation results of point clouds from mobile LiDAR devices, we propose an optimized segmentation method based on Scanline Continuity Constraint (SLCC) in this work. Unlike conventional scanline-based segmentation methods, SLCC clusters scanlines using the continuity constraints in terms of the distance as well as the direction of two consecutive points. In addition, scanline clusters are agglomerated not only into primitive geometrical shapes but also irregular shapes. Another downside to existing segmentation methods is that they are not capable of incremental processing. This causes unnecessary memory and time consumption for applications that require frame-wise segmentation or when new point clouds are added. In order to address this, we propose an incremental scheme—the Incremental Recursive Segmentation (IRIS), that can be easily applied to any segmentation method. IRIS is achieved by combining the segments of newly added point clouds and the previously segmented results. Furthermore, as an example application, we construct a processing pipeline consisting of plane fitting and surface reconstruction using the segmentation results. Finally, we evaluate the proposed methods on three datasets acquired from a handheld Velodyne HDL-32E LiDAR device. The experimental results verify the efficiency of IRIS for any segmentation method and the advantages of SLCC for processing mobile LiDAR data.

Highlights

Segmentation—the first step in most of these processes—is a very important and indispensable operation.There exist many segmentation algorithms, such as Region Growing [3] and Random Sample Consensus (RANSAC) [4].two main challenges are encountered when these are applied to LiDAR data
To address the second challenge, we propose a solution inspired by the frame-wise processing of point clouds in autonomous driving and robotics applications
We propose a novel segmentation method that is able to overcome the sparsity and non-uniformity of a point cloud from mobile LiDAR devices

Summary

Introduction

Segmentation—the first step in most of these processes—is a very important and indispensable operation.There exist many segmentation algorithms, such as Region Growing [3] and RANSAC [4].two main challenges are encountered when these are applied to LiDAR data. There exist many segmentation algorithms, such as Region Growing [3] and RANSAC [4]. 2016, 8, 967 arises because of the sparsity and non-uniformity of the point cloud acquired by mobile LiDAR devices when compared with the point cloud generated by depth sensors or static LiDAR devices. Sparsity and non-uniformity greatly increase the rate of over-segmentation and miss-segmentation. The second issue is associated with the ability to process the point cloud incrementally. The segmentation time increases drastically as the scale of the point cloud increases. The time-consuming computation has to be performed again if a new point cloud is added. This can be avoided if incremental segmentation is available

Objectives

Results

Conclusion