DSNRNet: A Network of LiDAR Data Denoising by Differential Stability in Underground Mine

Some sprinklers in underground mines spray water mist to lessen the amount of dust in the air. But water mist causes LiDAR to provide inaccurate point measurements, which are referred to as water mist noise that undermines the effectiveness of LiDAR-based localization and object recognition. Therefo...

Full description

Saved in:
Bibliographic Details
Main Authors: Meng, Dejiang, Tian, Bin, Pan, Ziyu, Wang, Junhui, Chen, Long
Format: Conference Proceeding
Language:English
Subjects:
Feature extraction
Laser radar
Point cloud compression
Rain
Semantic segmentation
Sensors
Snow
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Some sprinklers in underground mines spray water mist to lessen the amount of dust in the air. But water mist causes LiDAR to provide inaccurate point measurements, which are referred to as water mist noise that undermines the effectiveness of LiDAR-based localization and object recognition. Therefore, in order to reduce water mist noise, we have developed a new noise segmentation network that can operate on a CPU in real time-differential stability noise removal network (DSNRNet). This network consists of two sub-networks. The first sub-network is aimed at extracting differential stability features, so we named it the differential stability feature extraction network (sub-network 1). The second sub-network: a fully connected neural network (sub-network 2), is used to segment noise. To evaluate the DSNRNet's performance, we built a LiDAR semantic segmentation dataset of underground mines and run the DSNRNet in an Intel i7-11800H CPU. The experimental results demonstrate that this method is able to strike a better balance between speed (26.3 milliseconds) and accuracy (97.8 % ) compared with the other two most possible methods-DSOR (24.2 milliseconds, 3.2 % ) and WeatherNet (436.8ms, 98.5 % ).
ISSN:2153-0017
DOI:10.1109/ITSC57777.2023.10422095