Kusuma, Muhamad Nurramadhan (2021) Penggunaan CNN Untuk Re-identifikasi Orang Pada Lingkungan Multi Kamera. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.
Text
07211640000044-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2023. Download (3MB) | Request a copy |
Abstract
Re-identifikasi adalah suatu kegiatan mengidentifikasi sesuatu secara berulang ulang. Re-identifikasi memiliki arti proses mencocokan ulang data untuk menemukan hasil yang dicari. Re-identifikasi dapat dilakukan dengan cara menggunakan metode Deep learning dengan menggunakan arsitektur Convolutional Neural Network (CNN). Masih barunya peneranpan Re-identifikasi orang pada kamera Closed Circuit Television (CCTV) di indonesia. Jika sistem re-identifikasi ingin dikembangkan, maka harus menentukan model CNN manakah yang efektif dan efisien untuk diterapkan pada re-identifikasi pada orang. Sehingga dilakukan penelitian perbandingan berdasarkan tingkat performa antar mode CNN yang digunakan untuk re-identifikasi pada orang. Penelitian ini bertujuan untuk menemukan metode CNN yang efektif dan efisien dalam melakukan reidentifikasi pada orang. dengan membandingkan beberapa metode yang umum digunakan pada sistem re-identifikasi. Pada penelitian ini kami mengujikan beberapa model yaitu ResNet50,ResNet50 dengan Random Erasing, ResNet50 dengan Full Trick, ResNet50 dengan Random Erasing + Full Trick. DenseNet121, DenseNet121 dengan Random Erasing, DenseNet121 dengan Full Trick, DenseNet121 dengan Random Erasing + Full Trick. NAS, NAS dengan Random Erasing, NAS dengan Full Trick, NAS dengan Random Erasing + Full Trick. PCB, PCB dengan Random Erasing, PCB dengan Full Trick, PCB dengan Random Erasing + Full Trick. dari beberapa model tersebut kami mendapatkan nilai Rank1 sebesar 92%,Rank5 sebesar 97%,Rank10 sebesar 98% dan mAP sebesar 52% yang dihasilkan oleh model PCB. Model PCB adalah Model dengan nilai terbesar yang dihasilkan pada penelitian ini.
=====================================================================================================
Re-identification is an activity to identify something repeatedly. Re-identification means the process of re-matching data to find the results you are looking for. Re-identification can be done by using the Deep learning method using the Convolutional Neural Network (CNN) architecture. It is still the new implementation of Re-identification of people on Closed Circuit Television (CCTV) cameras in Indonesia. If a re-identification system is to be developed, it must determine which CNN model is effective and efficient to apply to the re-identification of people. So that a comparative study was conducted based on the level of performance between CNN modes used for re-identification of people. This study aims to find an effective and efficient CNN method in re-identifying people. by comparing several methods commonly used in re-identification systems. In this research, we tested several models, namely ResNet50, ResNet50 with Random Erasing, ResNet50 with Full Trick, ResNet50 with Random Erasing + Full Trick. DenseNet121, DenseNet121 with Random Erasing, DenseNet121 with Full Trick, Dense Net121 with Random Erasing + Full Trick. NAS, NAS with Random Erasing, NAS with Full Trick, NAS with Random Erasing + Full Trick. PCB, PCB with Random Erasing, PCB with Full Trick, PCB with Random Erasing + Full Trick. from some of these models we get Rank1 values of 92%, Rank5 of 97%, Rank10 of 98% and mAP of 52% generated by the PCB model. PCB model is the model with the largest value generated in this study
Item Type: | Thesis (Undergraduate) |
---|---|
Uncontrolled Keywords: | CCTV,Re-identifikasi,Deep Learning,CCTV,Reidentification,Deep Learning |
Subjects: | R Medicine > R Medicine (General) > R858 Deep Learning |
Divisions: | Faculty of Electrical Technology > Computer Engineering > 90243-(S1) Undergraduate Thesis |
Depositing User: | Muhamad Nurramadhan Kusuma |
Date Deposited: | 02 Sep 2021 08:58 |
Last Modified: | 02 Sep 2021 08:58 |
URI: | http://repository.its.ac.id/id/eprint/91101 |
Actions (login required)
View Item |