Setiawardhana, Setiawardhana (2021) Metode Penjejakan dan Blokade Bola Handal Menggunakan Kecerdasan Buatan Pada Robot Sepak Bola Beroda. Doctoral thesis, Institut Teknologi Sepuluh Nopember.
Preview |
Text
07111660010006-Dissertation.pdf - Accepted Version Download (2MB) | Preview |
Abstract
Penelitian tentang robot sepak bola beroda telah banyak dilakukan, tetapi belum ada yang spesifik meneliti sistem estimasi posisi bola dan navigasi robot penjaga gawang untuk memblokade bola di gawang. Penelitian ini terbagi menjadi beberapa tahapan penelitian. Penelitian pertama diawali dengan membuat sistem estimasi posisi bola visual dengan kamera fish-eye. Penelitian kedua adalah membuat sistem estimasi posisi bola dengan neural network dan kamera omnidirectional. Sistem dapat mentransformasi posisi bola dari bidang omni ke bidang datar atau kartesian. Hasil transformasi tersebut dikonversi ke area grid. Robot dapat mengetahui jarak bola sebenarnya di lapangan datar dimana banyak peneliti sebelumnya masih menggunakan kalibrasi manual dengan pengambilan data yang membutuhkan waktu proses lebih lama disaat perlombaan robot. Penelitian ketiga adalah membuat sistem navigasi Goalkeeper Robot Navigation (GKRN) untuk menggerakkan robot dalam memblokade bola. Sistem juga dapat memprediksi posisi bola dan waktu kedatangan di area tujuan menggunakan Modified Two-Layer Neural Network (MTLNN). MTLNN dibuat dengan modifikasi arsitektur neural network untuk memprediksi posisi dan waktu kedatangan bola. Metode GKRN dibuat dengan modifikasi Fuzzy Inferences System (FIS) pada Membership Function (MF) sesuai kebutuhan. Hasil untuk sistem pelacakan dan prediksi bola visual dengan area tersegmentasi mendapatkan keberhasilan untuk estimasi posisi bola sebesar 76.67% dan telah memenuhi capaian minimal untuk kriteria disain sebesar 60%. Hasil penelitian sistem estimasi posisi bola menggunakan neural network dengan kamera omni adalah akurasi estimasi posisi bola dengan nilai RMSE pada sumbu-x sebesar 10.3 cm dan RMSE pada sumbu-y sebesar 29,3 cm dengan simpul tersembunyi berjumlah 70. Hasil penelitian sistem navigasi dengan algoritma MTLNN menunjukkan peningkatan kemampuan prediksi hingga sekitar 20 kali lebih baik dari metode sebelumnya untuk posisi bola dan sekitar 4 kali lebih baik dari metode sebelumnya untuk waktu kedatangan, serta algoritma GKRN memiliki keberhasilan 90% dalam blokade bola.
=====================================================================================================
There have been many studies on wheeled soccer robots, but no one has specifically developed a ball position estimation system and goalkeeping robot navigation to block the ball in the goal. This research is divided into several stages of research. The research begins with creating a visual ball position estimation system using a fish-eye camera. The second research is to make a ball position estimation system with a neural network with an omnidirectional camera. The system can transform the position of the ball from an omni plane to a flat or cartesian plane. The result of the transformation is converted to a grid area. Robots can find out the actual distance of the ball on a flat field where many previous researchers still used manual calibration with data retrieval which required a longer processing time during a robot race. The third research is to create a Goalkeeper Robot Navigation (GKRN) navigation system to move the robot to block the ball. The system can also predict ball position and arrival time at the destination area using a Modified Two-Layer Neural Network (MTLNN). MTLNN is made with a modified neural network architecture to predict the position and time of arrival of the ball. The GKRN method is made by modifying the Fuzzy Inferences System (FIS) on the Membership Function (MF) as needed. The results for the visual ball tracking and prediction system with segmented areas have the success for estimating the ball position by 76.67% and have met the minimum success achievement above 60%. This result is the initial stage of research for the development of research with an omnidirectional camera. The results of further research on the ball position estimation system using a neural network with an omni camera are the accuracy of estimating the position of the ball with an RMSE value on the x-axis of 10.3 cm and an RMSE on the y-axis of 29.3 cm with a hidden node of 70. The MTLNN algorithm shows an increase in predictive ability up to about 20 times better than the previous method for ball position and about 4 times better than the previous method for arrival time, and the GKRN algorithm has 90% success in ball blockade.
| Item Type: | Thesis (Doctoral) |
|---|---|
| Uncontrolled Keywords: | estimasi posisi bola, prediksi posisi bola, robot penjaga gawang, transformasi gambar omni ke kartesian, algoritma grid, jaringan syaraf tiruan, blokade bola, MTLNN, GKRN. ======================================================== ball position estimation, ball position prediction, goalkeeping robot, omni to cartesian image transformation, grid algorithm, artificial neural network, ball blockade, MTLNN, GKRN. |
| Subjects: | Q Science > QA Mathematics > QA336 Artificial Intelligence T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK3070 Automatic control T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20001-(S3) PhD Thesis |
| Depositing User: | SETIAWARDHANA SETIAWARDHANA |
| Date Deposited: | 02 Mar 2021 02:40 |
| Last Modified: | 27 Sep 2024 03:30 |
| URI: | http://repository.its.ac.id/id/eprint/83067 |
Actions (login required)
 |
View Item |