|
A Comparative Study on Vehicle Physical Appearance Identification using Transfer Learning Methods |
|---|---|
| รหัสดีโอไอ | |
| Creator | Kahabodee Prakobchat |
| Title | A Comparative Study on Vehicle Physical Appearance Identification using Transfer Learning Methods |
| Contributor | Kwankamon Dittakan, Salang Musikasuwan |
| Publisher | Faculty of Engineering Mahasasakham University |
| Publication Year | 2566 |
| Journal Title | Engineering Access |
| Journal Vol. | 9 |
| Journal No. | 2 |
| Page no. | 154-167 |
| Keyword | Vehicle appearance detection, deep learning, object detection, transfer learning, convolutional neural network |
| URL Website | https://ph02.tci-thaijo.org/index.php/mijet/index |
| Website title | THAIJO Engineering Access |
| ISSN | 2730-4175 |
| Abstract | Traffic rule violations by drivers are a significant global concern, particularly in urban areas, as they contribute to increasing traffic accidents. This study proposed a novel approach to identifying vehicles involved in such violations by building prediction models using image processing and machine learning techniques. The research focused on three key vehicle characteristics: type, colour, and brand. The study employed a transfer learning mechanism as the machine learning method to generate the prediction models. The results revealed that the YOLO V8 achieved the highest accuracy in predicting vehicle type and colour, with an accuracy of 98.9% and 93.2%, respectively. Comparatively, YOLO V7, V6, V5, V4, and V3 achieved lower accuracies. In terms of predicting vehicle brand, the YOLO V8 achieved an accuracy of 89.8%, surpassing the accuracies of the YOLO V7, V6, V5, V4, and V3. These findings demonstrated the potential of image processing and machine learning techniques in accurately identifying vehicles involved in traffic violations and highlighted the opportunity to develop effective strategies to reduce the number of traffic accidents caused by rule violations. This research has significant implications for enhancing road safety and promoting advanced technologies to address real-world problems. |
