| Title |
Design and Implementation of a Portable License Plate Recognition System Using Edge AI |
| Authors |
이재희(Jae-Hee Lee) ; 박시형(Si-Hyung Park) ; 김지섭(Ji-Sub Kim) ; 서창진(Chang-Jin Seo) |
| DOI |
https://doi.org/10.5370/KIEE.2026.75.8.1965 |
| Keywords |
Coral Dual Edge TPU; YOLOv8n; EfficientDet-Lite0; LPRNet |
| Abstract |
This study presents the design and implementation of an advanced, real-time portable License Plate Recognition (LPR) system optimized for edge computing environments using the Raspberry Pi 5. To ensure high-speed processing and enhanced accuracy, a comparative performance analysis was conducted on AI accelerators, specifically the Coral USB Single TPU and M.2 Dual Edge TPU. For detection, YOLOv8n and EfficientDet-Lite0 were evaluated, while LPRNet was adopted for recognition. To optimize these models for edge deployment, transfer learning and integer quantization were strategically applied. Results revealed that EfficientDet- Lite0 outperformed YOLOv8n by a margin of 0.16 in mAP@50-95. Furthermore, by leveraging transfer learning and model format conversion, the LPRNet model size was reduced to one-tenth of its original footprint, achieving significant architectural efficiency. Robust real-time performance was successfully validated by integrating the Coral M.2 Dual Edge TPU. |