Klasifikasi Social Distancing Pada Data Video Menggunakan Metode CNN Dengan Arsitektur YOLO

Sudrajad, Bagas Yanuar (2021) Klasifikasi Social Distancing Pada Data Video Menggunakan Metode CNN Dengan Arsitektur YOLO. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 05111740000074-Undergraduate_Thesis.pdf] Text
05111740000074-Undergraduate_Thesis.pdf - Accepted Version
Restricted to Repository staff only until 1 October 2023.

Download (6MB) | Request a copy

Abstract

Social distancing, disebut juga physical distancing atau menjaga jarak, adalah tindakan menghindari kerumunan dan menjaga jarak fisik antar orang. Studi terkini menunjukkan menjaga jarak merupakan salah satu tindakan yang efektif untuk menghambat laju penyebaran penyakit menular seperti COVID-19. Akan tetapi, menjaga jarak sering kali masih tidak diterapkan karena kurangnya kesadaran dari masyarakat atau bahkan karena tempat yang ada terlalu penuh sehingga jaga jarak tidak dapat dilaksanakan. Karena itu, sistem yang dapat mendeteksi manusia secara otomatis dan melakukan klasifikasi social distancing akan memudahkan pengawasan dan pelaksanaan social distancing.
Tujuan tugas akhir ini adalah mengembangkan klasifikasi social distancing menggunakan metode You Only Look Once (YOLO). YOLO merupakan salah satu metode deteksi objek berbasis Convolutional Neural Network (CNN) yang terkenal dengan kecepatannya. Klasifikasi social distancing dilakukan dengan melakukan deteksi manusia menggunakan YOLO, mengambil titik tengah bawah bounding box hasil deteksi sebagai representasi posisi manusia di bidang gambar, melakukan transformasi posisi-posisi tersebut menggunakan matriks hasil kalibrasi kamera, mengukur jarak euclidean antara posisi yang ditransformasikan dan melakukan klasifikasi berdasarkan apakah suatu jarak melanggar ambang batas yang telah ditetapkan.
Dataset yang dipakai untuk pelatihan adalah data WiderPerson. Sebagian data WiderPerson dipakai untuk pelatihan model dan dilakukan dengan berbagai skenario, yaitu variasi arsitektur YOLO, variasi learning rate, dan variasi ukuran batch. Model terbaik didapatkan dari pelatihan dengan arsitektur YOLOv4, learning rate 0.001 dan ukuran batch 16. Model inilah yang selanjutnya akan digunakan untuk melakukan pengujian. Dari pengujian deteksi manusia, didapatkan rata-rata nilai mAP sebesar 82,91%. Dari pengujian klasifikasi social distancing, didapatkan rata-rata nilai akurasi sebesar 94,83%, rata-rata nilai recall sebesar 97,06%, rata-rata nilai precision sebesar 96,06%, dan rata-rata nilai F1 score sebesar 96,49%.
=========================================================
Social distancing, also known as physical distancing, is the act of avoiding crowds and maintaining physical distance between people. Recent studies have shown that social distancing is one of the most effective measures to slow the spread of infectious diseases such as COVID-19. However, social distancing is often not implemented due to lack of awareness from the community or even because the space is too crowded so that social distancing cannot be implemented. Therefore, a system that can detect humans automatically and perform social distancing classifications will facilitate the supervision and implementation of social distancing.
The purpose of this final project is to develop a social distancing classification using You Only Look Once (YOLO) method. YOLO is an object detection method that is based on Convolutional Neural Network (CNN) which is famous for its speed. Social distancing classification is carried out by detecting humans using YOLO, taking the lower midpoint of the bounding box as a representation of the human’s position in the image, transforming these positions using a camera calibration matrix, measure the euclidean distance between the transformed positions and perform a classification based on whether a distance violates a predetermined threshold.
The dataset used for training is the WiderPerson data. Some of the WiderPerson data is used for model training, which is performed with various YOLO architectures, learning rates, and batch sizes. The best model was obtained from training with YOLOv4 architecture, learning rate of 0.001 and batch size of 16. This model will then be used for testing. From human detection testing, the average mAP value of 82.91% was obtained. From the social distancing classification testing, an average accuracy value of 94.83%, an average recall value of 97.06%, an average precision value of 96.06, and an average F1 score of 96.49% were obtained.

Item Type: Thesis (Undergraduate)
Uncontrolled Keywords: Social Distancing, Convolutional Neural Network, YOLO, Deteksi Manusia, Social distancing, Convolutional Neural Network, YOLO, People Detection.
Subjects: Q Science > QA Mathematics > QA336 Artificial Intelligence
R Medicine > R Medicine (General) > R858 Deep Learning
R Medicine > RA Public aspects of medicine > RA644.C67 COVID-19 (Disease)
Divisions: Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User: Bagas Yanuar Sudrajad
Date Deposited: 15 Aug 2021 04:08
Last Modified: 15 Aug 2021 04:08
URI: http://repository.its.ac.id/id/eprint/86739

Actions (login required)

View Item View Item