Maulana, Hendra (2018) Deteksi Lokasi Plat Nomor Kendaraan Menggunakan Metode Maximally Stable Extremal Regions Dan Harris Corner. Masters thesis, Institut Teknologi Sepuluh Nopember.
Preview |
Text
5115201043 - Master_Thesis.pdf - Accepted Version Download (7MB) | Preview |
Abstract
Beberapa penelitian metode pengenalan plat telah menunjukkan kinerja menjanjikan, namun beberapa metode mungkin gagal dalam situasi yang lebih kompleks karena kompleksitas seperti variasi posisi dan orientasi plat, berbagai latar belakang, dan benda-benda non-plat. Untuk efisiensi pencocokan visual yang lebih tinggi, beberapa detektor keypoint cepat dan deskripsi yang sesuai telah dilakukan penelitian, seperti seperti fitur FAST, SURF, BRISK, Harris Corner. Dan ada juga fitur Maximally Stable Extremal Regions (MSER) yang metode pencarian keypoint berdasarkan extremal regionnya. Metode MSER telah diidentifikasi sebagai salah satu detektor wilayah terbaik karena ketahanannya terhadap perubahan sudut pandang, skala, dan pencahayaan, serta sensitif terhadap citra yang kabur.
Deteksi lokasi plat nomor kendaraan dengan metode deteksi sudut Harris Corner mampu mendeteksi secara baik pada file gambar dengan kondisi pencahayaan yang beragam, hal ini dikarenakan oleh latar belakang yang relatif terfokus pada sudut. Namun untuk beberapa gambar plat nomor yang memantulkan sinar matahari masih sulit dideteksi. Oleh sebab itu penelitian ini akan menggabungkan metode Maximally Stable Extremal Regions (MSER) dan metode Harris Corner pada tahap ekstraksi fitur untuk deteksi lokasi plat nomor kendaraan. Penggunaan metode MSER diharapkan dapat memperbaiki kekurangan metode Harris Corner pada gambar plat nomor yang memantulkan sinar matahari dan dapat mengurangi area yang akan dicocokan dengan template berdasarkan deteksi corner point teks. Ekstraksi dilakukan terhadap 80 dataset mobil dengan plat nomor kendaraan standar Indonesia. Proses ekstraksi fitur metode MSER digunakan untuk mendeteksi kandidat area teks, selanjutnya dilakukan ekstraksi fitur Harris Corner untuk mendeteksi corner points teks dan akan dicocokan dengan template. Hasil pengujian menunjukkan nilai akurasi sebesar 98,85% dengan rata-rata waktu komputasi 32,96 detik. Sedangkan nilai presisi dan recall masing-masing sebesar 67,61 untuk presisi, dan 79,66 untuk recall.
=============================================================================================================
Several studies of plate recognition methods have shown promising performance, but some methods may fail in more complex situations due to complexity such as position variation and plate orientation, various backgrounds, and non-plate objects. For higher visual matching efficiency, some fast keypoint detectors and corresponding descriptions have been undertaken, such as features of FAST, SURF, BRISK, Harris Corner. And there is also MSER feature that search method keypoint based on extremal region. Detection of vehicle number plate locations with Harris Corner angle detection method is able to detect well on image files with various lighting conditions, this is because of the relatively focused theoretical background at the angle. But for some drawing plate numbers that reflect sunlight is still difficult to detect.
MSER method has been identified as one of the best regional detectors because its resistance to changes in viewing angle, scale, and illumination, sensitive to blurry imagery. MSER is very efficient to detect characters with half or fully enclosed areas, especially on hole characters, such as 0, 6, 8, 9, A, B, D, P, Q. Therefore, this research will incorporate Maximally Stable Extremal Regions (MSER) method and Harris Corner method at feature extraction stage for detection of vehicle license plate location. The use of the MSER method is expected to improve the deficiency of the Harris Corner method on the number plate image reflecting sunlight and can reduce the area to be matched with the template based on the detection of the text corner. The extraction was performed on 80 car datasets with Indonesian standard vehicle number plates. Feature extraction process of MSER method is used to detect candidate text area, then extraction feature Harris Corner to detect corner points text and will be matched with template. The test results show an accuracy of 98.85% with an average computational time of 32.96 seconds. While the value of precission and recall respectively 67,61 for precision, and 79,66 for recall.
| Item Type: | Thesis (Masters) |
|---|---|
| Additional Information: | RTIf 006.4 Mau d |
| Uncontrolled Keywords: | Feature Extraction, Extremal Region, Corner Detection, Harris Corner, MSER |
| Subjects: | T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques |
| Divisions: | Faculty of Information and Communication Technology > Informatics > 55101-(S2) Master Thesis |
| Depositing User: | Maulana Hendra |
| Date Deposited: | 24 Apr 2018 02:37 |
| Last Modified: | 10 Aug 2020 07:35 |
| URI: | http://repository.its.ac.id/id/eprint/50967 |
Actions (login required)
 |
View Item |