Deteksi Dan Perhitungan Orang Berbasis Deep Learning Menggunakan Kamera Drone

Marcellinus, Samuel (2021) Deteksi Dan Perhitungan Orang Berbasis Deep Learning Menggunakan Kamera Drone. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.

Text 05111740000134-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until October 2023. Download (10MB) | Request a copy

Abstract

Keadaan Indonesia saat ini sedang mengalami kondisi
yang kurang baik disebabkan oleh COVID-19. Pemerintah sudah memberlakukan beberapa protokol kesehatan seperti melarang adanya kerumunan. Meskipun demikian, tetap masih banyak masyarakat Indonesia yang berkerumun dan mengabaikan protokol kesehatan tersebut. Mengenai kerumunan, penghitungan orang pada kerumunan secara cepat merupakan salah satu kemampuan yang sangat dicari saat ini. Terutama dalam hal keamanan dan perencanaan publik, ini dianggap sangat penting. Dalam pandemi COVID-19 seperti ini, penghitungan orang pada kerumunan dapat menjadi informasi yang berharga bagi penegak hukum.

Oleh karena itu, diusulkan sebuah metode berbasis Deep
Learning untuk membantu memecahkan masalah di atas.
Objektifnya adalah mendeteksi semua orang dalam kerumunan, mengukur orang yang terdeteksi dengan bounding box yang tepat, lalu menghitung jumlah orang di kerumunan tersebut. Metode ini bernama LSC-CNN (Locate, Size, and Count Convolutional Neural Network). LSC-CNN menggunakan multi-column architecture dengan top-down feature modulation untuk mendeteksi orang lebih baik pada berbagai macam resolusi. Yang menarik, metode yang
diusulkan ini dapat memberikan ukuran bounding box yang tepat dengan menggunakan anotasi kepala saja.

LSC-CNN menggunakan beberapa dataset untuk training seperti ShanghaiTech Part A, ShanghaiTech Part B, UCF-QNRF, dan 20 image tambahan. 20 Image tambahan merupakan dataset yang berisi gambar-gambar dari Indonesia dengan karakteristik terdapat orang berhijab. LSC-CNN memiliki skor yang sangat baik saat testing terhadap dataset ShanghaiTech Part A dengan nilai MAE 66,472 dan MSE 117,014.
=====================================================================================================
Indonesia is currently experiencing an unfavorable
condition due to COVID-19. The Indonesian government has
implemented several health protocols such as banning crowds. Nevertheless, there are still many Indonesian people who congregate and ignore the health protocol. Regarding crowd, quick counting of people in a crowd is one of the most sought after skills today. Especially in terms of security and public planning, this is considered very important. In a COVID-19 pandemic like this, crowd counting can be valuable information for law enforcement.

Therefore, a deep learning-based method is proposed to
help solve the above problem. The objective is to detect everyone in the crowd, measure the detected people with the right bounding box, then count the number of people in the crowd. This method is called LSC-CNN (Locate, Size, and Count Convolutional Neural Network). LSC-CNN uses a multi-column architecture with top�down feature modulation to better detect people at various resolutions. Interestingly, this proposed method can provide the exact size of the bounding box using only the head annotation.

LSC-CNN uses several datasets for training such as ShanghaiTech Part A, ShanghaiTech Part B, UCF-QNRF, and 20 additional images. The 20 additional images consists image from Indonesia with the characteristics there are people using hijab. LSC-CNN has a very good score when testing the ShanghaiTech Part A dataset with MAE value of 66,472 and MSE of 117,014.

Item Type:	Thesis (Undergraduate)
Uncontrolled Keywords:	COVID, Crowd Counting, Deep Learning, People Detection, Convolutional Neural Network.
Subjects:	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:	Samuel Marcellinus
Date Deposited:	11 Aug 2021 06:31
Last Modified:	11 Aug 2021 06:31
URI:	http://repository.its.ac.id/id/eprint/85425

Actions (login required)

View Item