Open AccessProcessing time increasement of non-rice object detection based on YOLOv3-tiny using Movidius NCS 2 on Raspberry Pi
Author(s) -
Nova Eka Budiyanta,
Catherine Olivia Sereati,
Ferry Rippun Gideon Manalu
Publication year - 2022
Publication title -
bulletin of electrical engineering and informatics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.251
H-Index - 12
ISSN - 2302-9285
DOI - 10.11591/eei.v11i2.3483
Subject(s) - raspberry pi , pixel , computer science , artificial intelligence , object (grammar) , computer graphics (images) , object detection , image processing , computer vision , computer hardware , embedded system , pattern recognition (psychology) , image (mathematics) , internet of things
This study aims to increase the processing time of detecting non-rice objects based on the you only look once v3-tiny (YOLOv3-tiny) model. The system was developed on the Raspberry Pi 4 embedded system with the Movidius neural compute stick 2 (NCS 2) implementation approach. Data object in the form of gravel on a bunch of rice in the video. The video data was obtained using a webcam with a resolution of 1920 x 1080 pixels with a total of 2759 frames. From the test results, frames per second (FPS) have increased by 1.27x in the Movidius NCS 2 implementation compared to processing using the central processing unit (CPU) from the Raspberry Pi 4. The object detection processing on video data is complete at 1871.408 seconds with 1.474 FPS using the CPU from the Raspberry Pi 4 and finished at 1477.141 seconds with 1.868 FPS using Movidius NCS 2. From these differences, it can be seen that the application of Movidius NCS 2 succeeded in increasing the object detection processing in this study by 26.69% with the YOLOv3-tiny model approach on the Raspberry Pi 4 embedded system.