A DEEP LEARNING APPROACH TO DETECT SMOKING MOTORCYCLIST USING TENSORFLOW

SHODRI, MUHAMAD (2020) A DEEP LEARNING APPROACH TO DETECT SMOKING MOTORCYCLIST USING TENSORFLOW. S1 thesis, Universitas Mercu Buana Jakarta.

Text (JURNAL MAHASISWA) Jurnal_MuhamadShodri_41516010152.pdf Restricted to Registered users only Download (1MB)

Abstract

Smoking while riding a motorcycle is strictly prohibited. Smoking and riding will defect the focus of the rider when it is done simultaneously. Besides, the ash from the cigarette will spread and fly to other motorcyclists. As a result, the chance of an accident will be even higher. The government has made regulations to prohibit smoking from riding a motorcycle. However, the regulation is still done manually by monitoring the bypassing motorcyclist. Therefore, this study aims to detect motorcyclist who is smoking automatically and minimize the risk level of accidents on the highway. This study uses the Faster R-CNN as a detector and the framework is TensorFlow Object Detection API. The total of data in this study are 967 images with the category of a motorcyclist who is smoking (smoke) and motorcyclist who is not smoking (not smoke). The experimental results of 15 mAP scores that represents 67k steps training are 0.81364. From the obtained result, Faster R-CNN can detect and distinguish between the categories of smoke and not smoke. Keywords: Faster R-CNN, Tensorflow, Object detection, Smoking, Smoking motorcyclist

Actions (login required)

View Item