Aplikasi mendeteksi gol dengan metode Convolutional Neural Network dan Transfer Learning = Application for goal detection with Convolutional Neural Network and Transfer Learning methods

Limando, Rovario Khogus (2023) Aplikasi mendeteksi gol dengan metode Convolutional Neural Network dan Transfer Learning = Application for goal detection with Convolutional Neural Network and Transfer Learning methods. Bachelor thesis, Universitas Pelita Harapan.

Preview	Text (Title) Title.pdf.pdf Available under License Creative Commons Attribution Non-commercial Share Alike. Download (25kB) | Preview
Preview	Text (Abstract) Abstract.pdf.pdf Available under License Creative Commons Attribution Non-commercial Share Alike. Download (226kB) | Preview
Preview	Text (ToC) ToC.pdf.pdf Available under License Creative Commons Attribution Non-commercial Share Alike. Download (490kB) | Preview
Preview	Text (Chapter1) Chapter1.pdf.pdf Available under License Creative Commons Attribution Non-commercial Share Alike. Download (655kB) | Preview
	Text (Chapter2) Chapter2.pdf.pdf Restricted to Registered users only Available under License Creative Commons Attribution Non-commercial Share Alike. Download (1MB)
	Text (Chapter3) Chapter3.pdf.pdf Restricted to Registered users only Available under License Creative Commons Attribution Non-commercial Share Alike. Download (2MB)
	Text (Chapter4) Chapter4.pdf.pdf Restricted to Registered users only Available under License Creative Commons Attribution Non-commercial Share Alike. Download (311kB)
	Text (Chapter5) Chapter5.pdf.pdf Restricted to Registered users only Available under License Creative Commons Attribution Non-commercial Share Alike. Download (138kB)
Preview	Text (Bibliography) Bibliography.pdf Available under License Creative Commons Attribution Non-commercial Share Alike. Download (443kB) | Preview
	Text (Appendices) Appendices.pdf Restricted to Repository staff only Available under License Creative Commons Attribution Non-commercial Share Alike. Download (899kB)

Abstract

Dalam ranah sepak bola, sebuah gol dipandang terjadi ketika bola telah sepenuhnya melewati garis gawang. Namun, terdapat insiden-insiden di mana gol yang seharusnya diakui sah malah dinyatakan tidak valid, dan sebaliknya. Dalam lingkungan sepak bola amatir, sering kali terjadi ketidakjelasan mengenai garis gawang. Garis gawang yang jelas masih memunculkan keraguan apakah gol yang dinyatakan valid atau tidak, keraguan semakin besar ketika garis gawang tidak terlihat sama sekali. Dengan menghadapi dilema ini, telah diperkenalkan sebuah inovasi teknologi yang dikenal sebagai goal line technology (GLT), yang bertujuan untuk membantu wasit dalam menetapkan keabsahan suatu gol. Namun, kendati manfaatnya yang nyata, implementasi teknologi GLT terhambat oleh biaya yang signifikan dalam pemasangannya dan juga biaya operasionalnya, menyebabkan sebagian operator liga cenderung enggan untuk mengadopsinya. Dalam tugas akhir ini dikembangkan sebuah aplikasi bergerak berbasis Android untuk mendeteksi kejadian gol sebagai alternatif dari GLT. Metode yang digunakan dalam pendeteksian kejadian gol adalah Machine Learning, dengan menerapkan algoritma Convolutional Neural Network, dan teknik Transfer Learning. Bersama dengan model pralatih (pretrained model) InceptionV3 data citra gawang dengan variasi posisi bola disekitar garis gawang dilatih menggunakan Transfer Learning untuk menghasilkan model baru. Model yang disimpan dalam format .tflite digunakan didalam prediksi keabsahan suatu gol berdasarkan citra dari tangkapan kamera smartphone. Pengujian model menggunakan data latihan dan data baru. Berdasarkan hasil prediksi aplikasi akan memberi notifikasi kepada pengguna saat terjadi gol. Akurasi dari pengujian model yang ada adalah sebesar 89%. Secara umum dapat disimpulkan bahwa model dapat digunakan oleh aplikasi untuk mengklasifikasi gol. Sekalipun demikian kehadiran wasit masih dibutuhkan untuk memverifikasi hasil prediksi dari aplikasi. / In the realm of football, a goal is seen as occurring when the ball has completely crossed the goal line. However, there have been incidents where goals that should have been recognized as valid were instead declared invalid, and vice versa. In the amateur football environment, there is often a lack of clarity regarding goal lines. If the visible goal line still raises doubts about whether the goal declared is valid or not, doubts are even greater when the goal line is not visible at all. By facing this dilemma, a technological innovation known as goal line technology (GLT) has been introduced, which aims to assist referees in determining the validity of a goal. However, despite its obvious benefits, the implementation of GLT technology is hampered by the significant costs of installation and operational costs, causing reluctance among some league operators to adopt it. In this final project, an Android-based mobile application was developed to detect goal events as an alternative to GLT. The method used in detecting goal events is Machine Learning, by applying the Convolutional Neural Network algorithm and Transfer Learning techniques. Together with the InceptionV3 pretrained model, the goal image data with variations in the position of the ball around the goal line was trained using Transfer Learning to produce a new model. The model saved in .tflite format is used to predict the validity of a goal based on images captured by a smartphone camera. Model testing using training data and new data. Based on the prediction results, the application will notify the user when a goal occurs. The accuracy of the existing model testing is 89%. In general, it can be concluded that the model can be used by applications to classify goals. However, in some case the presence of a referee is still needed to verify the prediction results from the application.

Actions (login required)

View Item