0099 Nrxwa

Trash Detection for Ocean Cleanup Robot
using Remotely Operated Vehicle (ROV)

Ryan Febriansyah, Soo Young Shin
Kumoh National Institute of Technology Gumi, South Korea

ryaan354@kumoh.ac.kr, wdragon@kumoh.ac.kr
Abstract
Human-made debris is frequently dumped into naturally occurring aquatic ecosystems, such as rivers
and oceans. Finding marine trash in rivers and oceans is crucial so that its effects on the ecosystem can be
recognized and minimized. The manual measurement of the amount of debris in the oceans is labor-
intensive, time-consuming, and has a small covering area. The underwater vehicle such as Remotely
Operated Vehicle (ROV) can be utilized for surveys. The unstructured character of the seabed terrain
poses the main obstacle to underwater image-based detection. Therefore, it is crucial to have a robust
feature identification system in circumstances like these. Hence, This study removes the need for trash
manual sampling by using an autonomous method using neural networks and computer vision models using
ROV. In addition, It is observed that our proposed network, achieved an overall accuracy of 93.9 % for
correct detections.
1. Introduction
Trash detection for ocean cleanup is an important 2. Proposed Scheme
problem that has received increasing attention in
recent years due to the negative impacts of marine In the proposed method for trash detection using an
litter on the environment and wildlife. One potential ROV and YOLO, the authors may have used a trash
solution to this problem is the use of remotely operated dataset to train the YOLO object detection model. A
vehicles (ROVs) equipped with trash detection dataset is a collection of data that is used to train,
capabilities to identify and collect trash in the ocean. validate, and test machine learning models.
One approach to trash detection for ROVs is to use the
Robot Operating System (ROS) and the You Only Look
Once (YOLO) algorithm. ROS is an open-source
software platform for robotics that provides a
framework for building, integrating, and running robotic
applications.
YOLO is a real-time object detection algorithm that
can be used to identify objects in images and videos. It
divides an input image into a grid of cells and uses Figure 1. System Block Diagram
convolutional neural networks (CNNs) to predict the
presence and location of objects within each cell. As shown in Fig. 1 to create the trash dataset, the
YOLO is fast and accurate, making it well-suited for authors have collected a set of images that contain
use in robotics applications. different types of trash and debris, such as plastic
Overall, the use of an ROV for trash detection in bottles, bags, and other common types of marine litter.
the ocean offers the potential to significantly improve They may have then annotated these images by drawing
the efficiency and effectiveness of ocean cleanup bounding boxes around each instance of trash and
efforts, helping to protect marine ecosystems and labeling each bounding box with the corresponding class
wildlife from the harmful effects of marine litter and (e.g., "plastic bottle," "plastic bag," etc.).
plastic pollution. Once the trash dataset has been created and
In this paper, the authors propose the use of an annotated, it can be used to train the YOLO object
ROV for trash detection in the ocean, with the aim of detection model. During training, the model is fed the
improving the efficiency and effectiveness of ocean annotated images and learns to predict the bounding
cleanup efforts. The ROV is equipped with various boxes and class probabilities for each instance of trash
sensors and cameras, which are used to detect and in the image.
classify different types of trash and debris. The The authors are publishing the bounding box results
authors also present a system for controlling the ROV
(i.e., the location and dimensions of the detected trash)
and processing the sensor data, as well as a set of
to ROS. This could allow the ROV to access the
experiments to evaluate the performance of the
bounding box information and use it to navigate and
proposed system.
interact with the detected trash. Publishing the bounding
box results to ROS may involve creating a message type
that defines the data structure for the bounding box
information (e.g., the x and y coordinates of the top-left
corner of the bounding box, the width and height of the
bounding box, and the class label). The YOLO object The authors may have also evaluated the
detection model would then generate bounding box data performance of their system in terms of its speed
and publish it to ROS as a message of this type. By and efficiency, for example, by measuring the
publishing the bounding box results to ROS, the authors frame rate at which the system is able to process
may have provided a way for the ROV to access and use and classify images. This is important in real-time
this information in real-time, enabling it to navigate and applications, such as trash detection using an
interact with the detected trash more effectively. This, in ROV, where the system needs to be able to
turn, may help to improve the efficiency and effectiveness operate quickly and efficiently. Overall, the
of ocean cleanup efforts. results of the authors' experiments suggest that
their proposed system for trash detection using
2.1. Robot Operating System (ROS)
an ROV and YOLO is effective and performs well
compared to other methods. This may be a
promising approach for improving the efficiency
and effectiveness of ocean cleanup efforts.
Figure 2. Result trash detection
As shown in Fig. 2, the data from bounding box that

contain the location x,y will be published into ROS Figure 4. Result trash detection
framework for further processing such as autonomous

navigation, Integrating ROS could provide a number of Acknowledgement
benefits, including improved communication, sensor
This research was supported by the
integration, and navigation and control capabilities. This
MSIT(Ministry of Science and ICT), Korea, under
could help to improve the efficiency and effectiveness of
the Grand Information Technology Research Center
the system in detecting and interacting with trash and
support program (IITP-2023-2020-0-01612)
debris in the ocean.
supervised by the IITP(Institute for Information &
3. Experiment communications Technology Planning & Evaluation)
and by Priority Research Centers Program through
It sounds like the authors of the paper have the National Research Foundation of Korea(NRF)
conducted experiments to evaluate the performance funded by the Ministry of Education, Science and
of their proposed system for trash detection using Technology”(2018R1A6A1A03024003).
YOLO. The result found that the system performed
the best among the different methods author
References
compared it to. There are several ways in which the
authors might have measured the performance of [1] Armitage S, Awty-Carroll K, Clewley D, Martinez-
their proposed system. As shown in Fig. 2, Some Vicente V. Detection and Classification of Floating
common evaluation metrics for object detection Plastic Litter Using a Vessel-Mounted Video Camera
and Deep Learning. Remote Sensing. 2022;
include accuracy and loss. These metrics measure
14(14):3425.
the accuracy of the object detection model, as well
[2] G. SaiTeja and P. S, "Design and Simulation of an
as its ability to detect all instances of the target Autonomous Surface Vehicle for Trash Collection
objects (i.e., trash and debris in this case). using ROS," 2021 IEEE 9th Region 10 Humanitarian
Technology Conference (R10-HTC), 2021, pp. 1-6,
[3] Martínez-Vicente V, Clark JR, Corradi P, Aliani S,
Arias M, Bochow M, Bonnery G, Cole M, Cózar A,
Donnelly R, Echevarría F, Galgani F, Garaba SP,
Goddijn-Murphy L, Lebreton L, Leslie HA, Lindeque
PK, Maximenko N, Martin-Lauzer F-R, Moller D,
Murphy P, Palombi L, Raimondi V, Reisser J, Romero
L, Simis SGH, Sterckx S, Thompson RC, Topouzelis
KN, van Sebille E, Veiga JM, Vethaak AD. Measuring
Figure 3. System Block Diagram Marine Plastic Debris from Space: Initial Assessment
of Observation Requirements. Remote Sensing. 2019;
11(20):2443.

0099 Nrxwa

Uploaded by

Copyright:

Available Formats

0099 Nrxwa

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

0099 Nrxwa

Uploaded by

Copyright:

Available Formats

Trash Detection for Ocean Cleanup Robot

using Remotely Operated Vehicle (ROV)

Kumoh National Institute of Technology Gumi, South Korea

Figure 2. Result trash detection

As shown in Fig. 2, the data from bounding box that

framework for further processing such as autonomous

You might also like