A machine learning-based automated vehicle classification implementation on toll system in Malaysia: a preliminary study
Congestion in toll plazas has prompted the exploration of various solutions, from infrastructure improvements to advanced technologies. Enhancing toll plaza infrastructure, such as constructing additional tollbooths and widening lanes while implementing electronic toll collection systems, has had so...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Book Chapter |
| Language: | English |
| Published: |
KICT Publishing
2024
|
| Subjects: | |
| Online Access: | http://irep.iium.edu.my/112233/1/112233_A%20machine%20learning-based%20automated%20vehicle.pdf http://irep.iium.edu.my/112233/ https://kulliyyah.iium.edu.my/kict/fyp-ebook-adict/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Congestion in toll plazas has prompted the exploration of various solutions, from infrastructure improvements to advanced technologies. Enhancing toll plaza infrastructure, such as constructing additional tollbooths and widening lanes while implementing electronic toll collection systems, has had some positive impacts. However, these existing measures have faced limitations in effectively addressing congestion. The use of mixed-mode lanes at the leftmost toll lanes still applied manual vehicle classification, which relies on human operators, but it has yet to sufficiently overcome congestion, given the diverse vehicle types and toll rates. This situation leads to human error and affects traffic flow. Although RFID (Radio frequency identification) technology has been widely adopted at only a few toll lanes, challenges in implementation have led to congestion issues due to insufficient infrastructure and reliability problems. Therefore, the outcome of this project is to develop the best model detector of automated real-time multiclass vehicle classification for all lanes in the toll plaza. This model input is extracted from a pre-trained 800 images, which consist of 6 classes of vehicles and their annotated XML file, respectively, for one stage detector: Faster Region-Convolutional Neural Network (Faster R-CNN), ResNet-50 and two-stage detectors; You Only Look Once (YOLO), YOLOv8 Darknet-53. The classification model performs well in YOLOv8 architecture with the highest mean average precision (MAP-50) of 95.0% and has a good performance measurement on loss function compared to Faster R-CNN architecture. |
|---|