| dc.description.abstract | In this era, humans heavily rely on vehicles as their primary means of mobility. Motor vehicles
are the most widely used type of transportation, valued highly and thus prone to theft. Combating this
crime is challenging. Although the police often recover stolen vehicles and evidence, returning them to
their rightful owners proves difficult when their identities have been erased. To address this issue, a
system is designed to integrate with vehicles, making it hard to alter their identities. This study proposes
a vehicle ownership verification system by matching vehicle registration documents (BPKB) with the
vehicles themselves. The system employs a Convolutional Neural Network algorithm, specifically
YOLOv5, to authenticate the BPKB. Training the deep learning model involved a dataset comprising
203 images of BPKB covers and 397 images of their identification sections. Additionally, the system
utilizes a microcontroller with Wi-Fi connectivity and a 5V buzzer as a responder. To apply the
appropriate YOLOv5 model for use in the identification process on the cover and identity sections, the
performance of the five YOLOv5 models, namely YOLOv5n, YOLOv5s, YOLOv5m, YOLOv5l, and
YOLOv5x, was tested. YOLOv5m was chosen for the cover, while YOLOv5n was selected for
identification. Testing included two scenarios: the impact of external lighting and the angle of the BPKB
placement. The lighting scenarios comprised four scenarios: daylight, bright (1050Lm), dim (200Lm),
and dark (0Lm). BPKB placement varied from 0˚ to 345˚ at intervals of 15˚ for five trials each. A total
of 11,520 experiments were conducted using six genuine and six fake BPKBs, yielding average
confidence scores above 90% for cover identification and above 80% for identification sections.
Overall, the verification tool performed well, meeting its intended design objectives. | en_US |
