Pengembangan Aplikasi Deteksi Objek Rokok dan Kegiatan Merokok Menggunakan Algoritma YOLOv3

Abstract

This research aims to create an application that can help supervise smoking activities using a deep learning algorithm, namely YOLOv3. Using 2 methods for development, the incremental for software development life cycle and black box testing. The dataset used image that collected from the internet sites and camera footage depicting of cigarette objects and smoking activities. The dataset was trained and tested using a split test application by separating the data into two datasets, for a separation is 85% for test and 15% for training. The model produces a mAP accuracy rate of 69.54% and averange loss of 0.189, with a cigarette detection percentage rate of 60% to 71% and 40% to 90% for smoking activities. For distances that can be detected in the range of 3 to 4 meters.