IMPLEMENTASI METODE SINGLE SHOT MULTIBOX DETECTOR (SSD) UNTUK OBJECT TRACKING SECARA REAL-TIME PADA SISTEM TOKO PINTAR BLIBLI MART

• Main Author: Alfarizi, Rizal (Author)
• Format: Book
• Published: 2020-08-28.
• Subjects: Thesis; NonPeerReviewed
• Online Access: Link Metadata
• Notes: http://repository.upi.edu/51506/1/S_KOM_1600807_Title.pdf; http://repository.upi.edu/51506/2/S_KOM_1600807_Chapter1.pdf; http://repository.upi.edu/51506/3/S_KOM_1600807_Chapter2.pdf; http://repository.upi.edu/51506/4/S_KOM_1600807_Chapter3.pdf; http://repository.upi.edu/51506/5/S_KOM_1600807_Chapter4.pdf; http://repository.upi.edu/51506/6/S_KOM_1600807_Chapter5.pdf; http://repository.upi.edu/51506/7/S_KOM_1600807_Appendix.pdf
• Keywords: L Education (General); QA75 Electronic computers. Computer science

Abstract-Self checkout system technology is being used by many retail industry. in development of self checkout system, RFID, barcode and QR are the most technology used for this system, although those technology require big cost because of those sensors and IoT technology. Amazon go develop self checkout system based on computer vision and sensor. Thus, customer get new experience in shopping cashierless. This research intend to build this system using Single Shot Multibox Detection (SSD) object detection method with Mobilnet as base net layer and action recognition for knowing which item is put or took away in rack using Convolutional Neural Network (CNN ) method and Motion History Image (MHI) as an input. SSD as object detection used to detect item in rack also as feature extraction for action recognition model for detecting person and then get bounding box to get person's ROI image and finaly those images is converted to MHI as an input for the model. Object detection has been trained by custom dataset and get mAP@0.5 83% for action recognition we built three models as comparison with MHI duration 20, 30 and 45, we get 90%,94%,93% model accuracy with custom dataset we collected.