Evaluasi Coral Dev Board Mini Dalam Pengolahan Citra Dari Sensor Secara Real-Time

Indrawan, Ridho Husni (2025) Evaluasi Coral Dev Board Mini Dalam Pengolahan Citra Dari Sensor Secara Real-Time. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5027211043-Undergraduate-Thesis.pdf - Accepted Version Restricted to Repository staff only Download (10MB)

Abstract

Kebutuhan akan pengolahan citra secara real-time di perangkat edge semakin meningkat seiring berkembangnya teknologi kecerdasan buatan dan Internet of Things (IoT). Penelitian ini bertujuan mengevaluasi performa Coral Dev Board Mini dalam pengolahan citra dari sensor video secara real-time. Coral Dev Board Mini merupakan perangkat edge computing yang dilengkapi dengan Edge TPU untuk mempercepat proses inferensi model AI. Sistem memanfaatkan kamera CCTV sebagai sumber data video streaming yang diproses menggunakan model deteksi objek TensorFlow Lite yang telah dioptimasi untuk Edge TPU, yaitu EfficientDet-Lite0, SSDLite MobileDet, dan SSD MobileNet V2. Evaluasi dilakukan dengan mengukur kecepatan inferensi, konsumsi CPU, RAM, serta kestabilan suhu perangkat. Hasil pengujian menunjukkan bahwa pada data gambar statis, rata-rata konsumsi CPU mencapai 46.23%, RAM 14.66%, dan suhu maksimum 71.23 ◦C, sedangkan pada video lokal, konsumsi CPU 46.73%, RAM 14.99%, dan suhu maksimum 71.21 ◦C, serta pada streaming RTSP, konsumsi CPU 47.57%, RAM 15.18%, dan suhu maksimum 72.26 ◦C. Kecepatan inferensi rata-rata pada input gambar mencapai 28 FPS untuk model SSD MobileNet V2, sedangkan pada video lokal dan streaming RTSP berkisar antara 13–17 FPS bergantung pada model yang digunakan. Berdasarkan pengujian tersebut, Coral Dev Board Mini terbukti mampu melakukan pengolahan citra secara efisien dan stabil dengan konsumsi daya rendah tanpa memerlukan pendinginan tambahan. Namun, terdapat beberapa kendala teknis terkait kestabilan koneksi RTSP dan keterbatasan buffer yang mempengaruhi latensi visual. Berdasarkan hasil evaluasi tersebut, Coral Dev Board Mini menunjukkan potensi baik untuk diimplementasikan dalam sistem pengolahan citra real-time berbasis edge AI, khususnya pada aplikasi yang memerlukan inferensi lokal tanpa bergantung pada komputasi awan.
==================================================================================================================================
The demand for real-time image processing on edge devices continues to increase along with the development of artificial intelligence and Internet of Things (IoT) technologies. This study aims to evaluate the performance of the Coral Dev Board Mini in real-time image processing from video sensor data. The Coral Dev Board Mini is an edge computing device equipped with an Edge TPU to accelerate AI model inference. The system utilizes CCTV cameras as streaming video sources processed using TensorFlow Lite object detection models optimized for Edge TPU, namely EfficientDet-Lite0, SSDLite MobileDet, and SSD MobileNet V2. The evaluation was carried out by measuring inference speed, CPU usage, RAM usage, and device temperature stability. The experimental results show that for static image data, the average CPU usage reached 46.23%, RAM 14.66%, and a maximum temperature of 71.23◦C. For local video, CPU usage was 46.73%, RAM 14.99%, and a maximum temperature of 71.21◦C. For RTSP streaming, CPU usage was 47.57%, RAM 15.18%, and a maximum temperature of 72.26◦C. The average inference speed for static images reached 28 FPS for the SSD MobileNet V2 model, while for local video and RTSP streaming it ranged between 13–17 FPS depending on the model used. Based on these tests, the Coral Dev Board Mini has been proven to perform image processing efficiently and stably with low power consumption without requiring additional cooling. However, some technical issues were identified related to RTSP connection stability and buffer limitations affecting visual latency. Based on these evaluation results, the Coral Dev Board Mini demonstrates good potential for implementation in real-time edge AIbased image processing systems, particularly for applications requiring local inference without reliance on cloud computing.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Coral Dev Board Mini, Edge TPU, Pengolahan Citra, Real-Time, Edge AI, TensorFlow Lite ; Coral Dev Board Mini, Edge TPU, Image Processing, Real-Time, Edge AI, TensorFlow Lite.
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information Technology > 59201-(S1) Undergraduate Thesis
Depositing User:	Ridho Husni Indrawan
Date Deposited:	01 Aug 2025 01:53
Last Modified:	01 Aug 2025 01:53
URI:	http://repository.its.ac.id/id/eprint/124293

Actions (login required)

View Item