Implementasi Human Detection pada Amlogic S905X CPU dengan Sistem Operasi Android

Tunggorono, Zuhairaja Musheera (2023) Implementasi Human Detection pada Amlogic S905X CPU dengan Sistem Operasi Android. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 05311940000033-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 September 2025. Download (4MB) | Request a copy

Abstract

Seiring dengan perkembangan teknologi, pengimplementasian teknologi Artificial Intelligence (AI) pada berbagai sektor di kehidupan manusia sangat diperlukan untuk menunjang berbagai keperluan, salah satunya adalah Human Detection. Pada implementasi Human Detection di perangkat ini perlu menggunakan library Tensorflow Lite agar dapat berjalan di perangkat dengan sumber daya terbatas, pada implementasi ini juga diperlukan model dengan arsitektur yang menggunakan Convolutional Neural Network (CNN). Dengan penelitian ini menghasilkan penggunaan teknologi Human Detection menggunakan kamera RTSP pada perangkat ZTE B680H V5 Android yang memiliki spesifikasi Amlogic S905X CPU menggunakan model-model deep learning dan TensorFlow Lite, deteksi manusia secara real-time dapat dilakukan secara akurat dan responsif. Hal ini memungkinkan penggunaan perangkat tersebut untuk berbagai keperluan seperti pengawasan dan keamanan di berbagai sektor, serta pengembangan sistem cerdas berbasis visi komputer. Dari hasil penelitian menghasilkan kesimpulan model MobileNet memiliki waktu inferensi yang jauh lebih cepat dibandingkan dengan model EfficientDet Lite namun untuk lingkungan dengan banyak benda seperti di lab komputer, EfficientDet Lite lebih baik dalam mendeteksi manusia dan dengan tingkat kepercayaan lebih tinggi. MobileNet V2 yang sudah di latih dengan sebanyak 93.328 foto dan test atau validasi sebanyak 23.311 memiliki mAP 0.5194 dan evaluasi total_loss 0.4789 menghasilkan waktu inferensi yang sedikit lebih lambat dari MobileNet V1 (450 ms) namun memiliki tingkat kepercayaan lebih tinggi terlebih pada objek yang dekat dengan kamera. Waktu eksekusi tertinggi dimiliki oleh Model EfficientDet-lite 2 dengan 2265 ms, diikuti oleh EfficientDet-lite 1 dengan 1508 ms. Sedangkan, waktu eksekusi terendah dimiliki oleh SSD MobileNet V1 dengan 450 ms, dan SSD MobileNet V2 Trained dengan 1215 ms. Namun, terkadang MobileNet V2 dapat menghasilkan waktu inferensi yang jauh lebih cepat, dipengaruhi oleh jenis kamera yang digunakan dan kecepatan internet
==================================================================================================================================
With the advancement of technology, Artificial Intelligence (AI) technology has become increasingly necessary in various aspects of human life, particularly in Human Detection. One of the increasingly popular applications of AI technology is Human Detection or real-time recognition of humans using cameras, especially in improving surveillance and security in various sectors such as industry, commerce and the public. In implementing Human Detection on this device, it is necessary to use the Tensorflow Lite library so that it can run on devices with limited resources. This implementation also requires a model with an architecture that uses a Convolutional Neural Network (CNN). This research results in the use of Human Detection technology using RTSP cameras on ZTE B680H V5 Android devices that have Amlogic S905X CPU specifications using deep learning and TensorFlow Lite models, real-time human detection can be carried out accurately and responsively. This allows the use of these devices for various purposes such as surveillance and security in various sectors, as well as the development of computer vision-based intelligent systems. From the results of the study it was concluded that the MobileNet model has a much faster inference time compared to the EfficientDet Lite model but for environments with many objects such as in a computer lab, EfficientDet Lite is better at detecting humans and with a higher level of confidence. MobileNet V2 which has been trained with 93,328 photos and 23,311 tests or validations has a mAP of 0.5194 and an evaluation of total_loss 0.4789 produces an inference time that is slightly slower than MobileNet V1 (450 ms) but has a higher level of confidence, especially on objects close to the camera. The highest execution time is owned by Model EfficientDet-lite 2 with 2265 ms, followed by EfficientDet-lite 1 with 1508 ms. Meanwhile, the lowest execution time was owned by SSD MobileNet V1 with 450 ms, and SSD MobileNet V2 Trained with 1215 ms. However, sometimes MobileNet V2 can produce much faster inference times, depending on the type of camera used and internet speed

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Human Detection, Tensorflow Lite, RTSP, CNN, Amlogic S905X
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.774.A53 Android R Medicine > R Medicine (General) > R858 Deep Learning T Technology > T Technology (General)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information Technology > 59201-(S1) Undergraduate Thesis
Depositing User:	Zuhairaja Musheera Tunggorono
Date Deposited:	25 Sep 2023 01:12
Last Modified:	25 Sep 2023 01:12
URI:	http://repository.its.ac.id/id/eprint/102133

Actions (login required)

View Item