Mulia, Muhammad Abdan (2024) Pengembangan Metode GHARLPR: Reparameterisasi Struktural Pada Deteksi Dan Pengenalan Plat Nomor. Masters thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 6025222006-Master_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6025222006-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2027. Download (4MB) | Request a copy |
Abstract
Pengenalan plat nomor kendaraan bermotor dari gambar dan video dengan CCTV telah masif diterapkan dalam bidang transportasi dan kota cerdas. Tantangan signifikan muncul dalam pengembangan metode untuk perangkat dengan sumber daya ringan dan memori terbatas. Arsitektur jaringan saraf tiruan yang efisien sering dioptimalkan untuk mengurangi Floating-point Operations Per Second (FLOP) atau jumlah parameter. Metrik ini tidak berkorelasi terhadap latensi pada perangkat terbatas. Pada penelitian ini diusulkan metode Generalized Hybrid Structural Reparameterization for Automatic License Plate Recognition (GHARLPR). Dilakukan reparameterisasi struktural untuk penurunan percabangan pada model, deteksi kendaraan dan plat nomor secara bersamaan. Modul usulan GHAR hanya menggunakan satuperkalian pada bagian bottleneck dengan N percabangan, kemudian digabungkan menjadi satu cabang selama mode inferensi. Reparameterisasi struktural yang ditingkatkan ini dapat meningkatkan fitur dari berbagai peta f itur dan meningkatkan pemanfaatan parameter serta kompleksitas dalam inferensi. Hasil penelitian ini sistem mampu mendeteksi dan mengenali pelat dengan akurat dan komputasi yang ringan. Penelitian ini diuji secara komprehensif dengan studi kasus rekaman video pelat dari berbagai kondisi dengan dataset Large Scale Video License Plate (LSV-LP) dengan hasil F1-Score 85,0 dengan kecepatan 4 millisecond pada inferensi pada penarapannya menggunakan Graphics Processing Units (GPU).
=================================================================================================================================
The recognition of motor vehicle license plates from images and video using CCTV has been widely applied in the fields of transportation and smart cities. Significant challenges arise in developing methods for devices with lightweight resources and limited memory. Efficient neural network architectures are often optimized to reduce Floatingpoint Operations Per Second (FLOP) or the number of parameters. However, these metrics do not correlate with latency on limited devices. This study proposes a method called Generalized Hybrid Structural Reparameterization for Automatic License Plate Recognition (GHARLPR). Structural reparameterization is performed to reduce branching in the model, enabling simultaneous detection of vehicles and license plates. The proposed GHAR module uses only one multiplication in the bottleneck section with N branches, which are then combined into a single branch during inference mode. This enhanced structural reparameterization can improve features from various f ature maps and increase the utilization of parameters and complexity in inference. The results of this study show that the system can accurately detect and recognize plates with lightweight computation. The research is comprehensively tested with a case study of video recordings of plates under various conditions using the Large Scale Video License Plate (LSV-LP) dataset, achieving an F1-Score of 85.0 with a speed of 4 milliseconds per inference when implemented using Graphics Processing Units (GPU).
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Automatic License Plate Recognition (ALPR), Deteksi dan Pengenalan Pelat Nomor, Scene Text Recoginition (STR), Automatic License Plate Recognition (ALPR), License Plate Detection, License Plate Recognition, Scene Text Recoginition (STR) |
| Subjects: | T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T58.62 Decision support systems |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis |
| Depositing User: | Muhammad Abdan Mulia |
| Date Deposited: | 23 Jan 2025 01:28 |
| Last Modified: | 23 Jan 2025 01:28 |
| URI: | http://repository.its.ac.id/id/eprint/116647 |
Actions (login required)
 |
View Item |