MULANA, AAN (2023) PERBANDINGAN SSD-MOBILENETV2 DENGAN SSD LITE- MOBILENETV2 MENGGUNAKAN RASPBERRY PI UNTUK KEAMANAN RUMAH SECARA REAL-TIME. S1 thesis, Fakultas Teknik Universitas Sultan Ageng Tirtayasa.
Text
Aan_Mulana_3332160067_Fulltext.pdf Download (3MB) |
|
Text
Aan_Mulana_3332160067_01.pdf Download (3MB) |
|
Text
Aan_Mulana_3332160067_02.pdf Download (1MB) |
|
Text
Aan_Mulana_3332160067_03.pdf Download (1MB) |
|
Text
Aan_Mulana_3332160067_04.pdf Download (1MB) |
|
Text
Aan_Mulana_3332160067_05.pdf Download (1MB) |
|
Text
Aan_Mulana_3332160067_Ref.pdf Download (1MB) |
|
Text
Aan_Mulana_3332160067_Lamp.pdf Download (1MB) |
|
Text
AAN_MULANA_3332160067_CP.pdf Download (14MB) |
Abstract
This research is designed to create a real-time home security system using Raspberry Pi. The system operates by detecting humans captured by the Pi camera, then processed by the Raspberry Pi. Processing is carried out by the Raspberry Pi with the assistance of the Tensorflow Lite framework, resulting in outputs distinguishing between known and unknown person. If the unknown person is detected, thelsystemlwill send the information to Telegramllinlthelformlof images and text. Detecting humans requires training data, which is accomplished using Google Colaboratory due to its super GPU that accelerates the training process and produces SSD-MobilenetV2 and SSD Lite-MobilenetV2 models. The result of SSD-MobilenetV2 model has an average accuracy value of 97.35% and a frame rate of 2.5 frames per second (fps). Meanwhile, the result of the SSD-Lite MobilenetV2 model has an average accuracy value of 96.72% and a frame rate of 5.6 fps. The optimal result for human detection model is SSD Lite-MobilenetV2, as it maintains a frame rate of 5.6 fps and its accuracy value is not significantly different from the SSD-MobilenetV2 model.
Item Type: | Thesis (S1) | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
Contributors: |
|
|||||||||
Additional Information: | Penelitian ini dirancang untuk membuat sistem keamanan rumah dengan Raspberry Pi secara real-time. Sistem ini bekerja dengan mendeteksi manusia yang ditangkap oleh Pi camera kemudian diproses oleh Raspberry Pi. Pemprosesan oleh Raspberry Pi dengan bantuan dari framework Tensorflow Lite sehingga menghasilkan output orang yang dikenal dan tidak dikenal. Apabila mendeteksi orang tidak dikenal maka akan mengirimkan informasi ke Telegram berupa gambar dan teks. Mendeteksi manusia diperlukan training data. Training data menggunakan Google Colaboratory karena terdapat super GPU yang mempercepat proses training dan menghasilkan model SSD-MobilenetV2 dan SSD Lite-MobilenetV2. Pengujian model SSD-MobilenetV2 memiliki nilai rata-rata akurasi sebesar 97,35% dan fps 2,5. Pengujian model SSD Lite-MobilenetV2 memiliki nilai rata-rata akurasi sebesar 96,72% dan fps 5,6. Model SSD Lite-MobilenetV2 merupakan model pengujian yang memiki hasil data paling maksimal dalam pendeteksian manusia karena memiliki nilai fps 5,6 dan nilai akurasi tidak berbeda jauh dari SSD-MobilenetV2. | |||||||||
Subjects: | Communication > Science Journalism T Technology > TK Electrical engineering. Electronics Nuclear engineering |
|||||||||
Divisions: | 03-Fakultas Teknik 03-Fakultas Teknik > 20201-Jurusan Teknik Elektro |
|||||||||
Depositing User: | Aan Mulana | |||||||||
Date Deposited: | 06 Oct 2023 14:51 | |||||||||
Last Modified: | 06 Oct 2023 14:51 | |||||||||
URI: | http://eprints.untirta.ac.id/id/eprint/30293 |
Actions (login required)
View Item |