Article
Spherically Stratified Point Projection: Feature Image
Generation for Object Classification Using 3D LiDAR Data
Chulhee Bae
1,†
, Yu-Cheol Lee
2,3,†
, Wonpil Yu
2
and Sejin Lee
1,
*
Citation: Bae, C.; Lee, Y.-C.; Yu, W.;
Lee, S. Spherically Stratified Point
Projection: Feature Image Generation
for Object Classification Using 3D
LiDAR Data. Sensors 2021, 21, 7860.
https://doi.org/10.3390/s21237860
Academic Editor: Subhas
Mukhopadhyay
Received: 5 October 2021
Accepted: 16 November 2021
Published: 25 November 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
1
Department of Mechanical Engineering, Kongju National University, Cheonan 31080, Korea;
bch3494@kongju.ac.kr
2
Artificial Intelligence Laboratory, ETRI, Daejeon 34129, Korea; yclee@etri.re.kr (Y.-C.L.); ywp@etri.re.kr (W.Y.)
3
Department of Computer Software, University of Science and Technology, Daejeon 34113, Korea
* Correspondence: sejiny3@kongju.ac.kr
† These authors contributed equally to this work.
Abstract:
Three-dimensional point clouds have been utilized and studied for the classification of
objects at the environmental level. While most existing studies, such as those in the field of computer
vision, have detected object type from the perspective of sensors, this study developed a specialized
strategy for object classification using LiDAR data points on the surface of the object. We propose
a method for generating a spherically stratified point projection (sP
2
) feature image that can be
applied to existing image-classification networks by performing pointwise classification based on a
3D point cloud using only LiDAR sensors data. The sP
2
’s main engine performs image generation
through spherical stratification, evidence collection, and channel integration. Spherical stratification
categorizes neighboring points into three layers according to distance ranges. Evidence collection
calculates the occupancy probability based on Bayes’ rule to project 3D points onto a two-dimensional
surface corresponding to each stratified layer. Channel integration generates sP
2
RGB images with
three evidence values representing short, medium, and long distances. Finally, the sP
2
images are
used as a trainable source for classifying the points into predefined semantic labels. Experimental
results indicated the effectiveness of the proposed sP
2
in classifying feature images generated using
the LeNet architecture.
Keywords: spherically stratified point project; feature image; semantic labeling; point cloud
1. Introduction
Object recognition in autonomous navigation relies on various deep-learning methods to
perform tasks such as detection and classification, which are being actively
investigated [1,2].
The safety of the path traveled by an autonomous vehicle and its ability to avoid obstacles
depend on the accurate classification of objects surrounding the vehicle [
3
]. Apart from au-
tonomous driving, object classification is necessary for various applications and represents
the basis for object recognition [4,5].
Owing to the accessibility of data, most object-classification methods use images
collected through vision sensors. A vision sensor, however, is greatly influenced by envi-
ronmental factors such as lighting and weather conditions. Consequently, unstable perfor-
mance may be observed in autonomous vehicles that rely only on image data, leading to
insufficient reliability for deployment in real scenarios [
6
,
7
]. Therefore, object recognition
based on LiDAR sensors, which are less sensitive to environmental factors than vision
sensors, must be studied together [8].
Various deep-learning methods have been applied for object classification using
only data from LiDAR sensors. The performance of such methods depends on distance
measurements from LiDAR sensors and the input data used for learning [
9
,
10
]. In fact, the
training performance of a deep-learning algorithm varies according to the input data, and
the quality and quantity of the input data can substantially affect the learning
results [11].
Sensors 2021, 21, 7860. https://doi.org/10.3390/s21237860 https://www.mdpi.com/journal/sensors
