Wisesa, Lukas Purba (2022) Deteksi Kerumunan untuk Pencegahan Penularan Virus Corona Berbasis Video Menggunakan Mask-RCNN. Masters thesis, Institut Teknologi Sepuluh Nopember.
Text
6022211057-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2024. Download (6MB) | Request a copy |
Abstract
Pandemi Covid-19 yang berlangsung dari tahun 2020 telah berlangsung selama 2 tahun. Walaupun banyak masyarakat telah menerima vaksin, pemerintah tetap menegaskan masyarakat untuk melakukan protokol kesehatan terutama di tempat umum. Mengikuti saran dari WHO, salah satu protokol kesehatan yang perlu diterapkan adalah menjaga jarak satu sama lain dengan jarak satu meter. Namun, dalam penerapan protokol kesehatan, tentunya banyak masyarakat yang lalai untuk menerapkan hal tersebut. Oleh karena itu, sebuah sistem deteksi yang dapat membantu memudahkan penerapan protokol kesehatan dirasa dibutuhkan.
Penelitian ini memanfaatkan algoritma Mask-RCNN yang digunakan untuk mendeteksi apakah terdapat kerumunan dalam suatu gambar yang diambil. Objek yang dideteksi pada penelitian ini dibagi menjadi tiga kelas yaitu : Crowd, Group, dan Person. Kelas Crowd merupakan kelas dimana terletak 4 orang atau lebih dalam radius 1 m, kelas group merupakan kelas dimana terdapat 2-3 orang dalam radius 1 m sedangkan kelas person merupakan kelas dimana hanya terdapat 1 orang dalam radius 1 m. Mask R-CNN mendeteksi objek dengan cara memberikan beberapa proposal objek yang kemudian diproses dengan layer konvolusi dan fully connected layer. Keluaran dari layer konvolusi adalah mask yang menunjukkan segmentasi gambar, sedangkan keluaran dari fully connected layers adalah memberikan bounding box serta menentukan nama kelas dari objek tersebut. Hasil dari penelitian ini adalah model yang dapat mendeteksi adanya kerumunan dengan akurasi
sebesar 88.79%.
===================================================================================================
The Covid-19 pandemic that started in 2020 has been going on for 2 years.
Although many people have received the vaccine, the government continues
to emphasize the public to carry out health protocols especially in public
places. Following the advice of WHO, one of the health protocols that need
to be implemented is to keep a meter distance from each other. However,
in implementing the health protocol, of course, many people tend to neglect
the physical distancing practice. Therefore, a detection system that can help
facilitate the implementation of health protocols especially physical distancing
is needed. Crowd class is a class where there are 4 or more people within a 1 m
radius, group class is a class where there are 2-3 people within a radius a meter
while the person class is a class where there is only 1 person within a meter
radius. This study utilizes the Mask-RCNN algorithm which is used to detect
whether there is a crowd in an image taken. The objects detected in this study
were divided into three classes, namely: Crowd, Group, and Person. The mask
R-CNN algorithm detects objects by providing several object proposals which are
then processed with the convolution layer and fully connected layer. The output
of the convolution layer is mask which shows the image segmentation, while the
output of fully connected layers is to give the bounding box and specify the class
name of the object. The result of this study is a model that can detect crowds
with an accuracy of 88.79%
Item Type: | Thesis (Masters) |
---|---|
Uncontrolled Keywords: | Mask R-CNN, Deteksi Kerumunan, deep learning, Covid-19, Mask R-CNN, Crowd Detection, Deep Learning, Covid-19. |
Subjects: | T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7882.P3 Pattern recognition systems |
Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis |
Depositing User: | Lukas Purba Wisesa |
Date Deposited: | 08 Jun 2022 04:43 |
Last Modified: | 08 Jun 2022 04:43 |
URI: | http://repository.its.ac.id/id/eprint/94900 |
Actions (login required)
View Item |