Penilaian Performa Pemain Sepak Bola Menggunakan Sistem Pelacakan Multi Objek Berbasis Yolov11

Syuhada, Ghifari Maaliki Syafa (2025) Penilaian Performa Pemain Sepak Bola Menggunakan Sistem Pelacakan Multi Objek Berbasis Yolov11. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211158-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Sepak bola merupakan salah satu olahraga paling populer di dunia, sehingga analisis performa pemain menjadi bidang yang semakin relevan untuk diteliti. Berbagai pendekatan berbasis sensor seperti GNSS, IMU, dan UWB telah digunakan untuk mengevaluasi performa pemain. Meskipun akurat, pendekatan ini memiliki keterbatasan seperti biaya yang tinggi, ketergantungan pada perangkat tambahan, dan keterbatasan akses terhadap data. Sebagai alternatif, pelacakan pemain berbasis citra video dengan pendekatan pelacakan multi objek berbasis deep learning menawarkan solusi yang lebih fleksibel tanpa perangkat tambahan. Namun, tantangan seperti oklusi, kemiripan seragam, kualitas video yang rendah, serta gerakan pemain yang cepat dan tidak terduga perlu diperhatikan dalam pengembangan sistem pelacakan. Oleh karena itu, dibutuhkan sistem yang mampu menangani kondisi non-linear, dinamis, dan kompleks yang umum terjadi dalam pertandingan sepak bola.
Penelitian ini mengimplementasikan sistem penilaian performa pemain berbasis model pralatih YOLOv11. Pelacakan dilakukan menggunakan algoritma ByteTrack dengan deteksi pemain dari YOLOv11 yang telah di fine tune menggunakan dataset SoccerNet-Tracking. Dua skema pengujian diterapkan, yaitu tanpa postprocessing dan dengan postprocessing menggunakan Global Tracklet Association (GTA). Estimasi matriks homografi dilakukan berdasarkan deteksi keypoint lapangan dari model YOLOv11 yang di fine tune menggunakan dataset Roboflow. Selanjutnya, posisi bounding box pemain ditransformasikan ke bidang dua dimensi melalui matriks homografi, dan digunakan untuk menghitung jarak serta kecepatan pemain berdasarkan euclidean distance.
Pengujian dilakukan pada klip uji dari test set SoccerNet-Tracking. Berdasarkan metrik HOTA, kombinasi ByteTrack dengan GTA menunjukkan performa pelacakan yang lebih baik dibandingkan ByteTrack saja. Namun, hasil proyeksi pemain masih menunjukkan ketidakstabilan, sehingga penilaian performa pemain belum sepenuhnya akurat. Hal ini menunjukkan bahwa penyempurnaan pada tahap proyeksi sangat penting agar evaluasi performa pemain dapat dilakukan secara lebih presisi dan konsisten.
===============================================================================================================================================
Football is one of the most popular sports in the world, making player performance analysis an increasingly relevant field of study. Various sensor-based approaches such as GNSS, IMU, and UWB have been used to evaluate player performance. Although accurate, these methods have limitations, including high costs, dependency on additional equipment, and restricted access to data. As an alternative, player tracing using video based multi object tracking powered by deep learning offers a more flexible soltion without requiring additional devices. However, challenges such as occlusion, similar player uniforms, low video quality, and unpredictable player movements must be addressed in the development of tracking systems. Therefore, a robust tracking system capable of handling nonlinear, dynamic and complex conditions typical infootball matches is needed.
This study implements a player performance evaluation system based on the pretrained YOLOv11 model. Player tracking is performed using ByteTrack algorithm with detections from YOLOv11, which as fine tuned using the SoccerNet-Tracking dataset. Two tracking schemes are evaluated: one without postprocessing and another with postprocessing using Global Tracklet Association (GTA). The homography matris is estimated based on keypoints detected by the YOLOv11 model fine tuned with a dataset from Roboflow. Subsequently, the bounding box positions of each player are transformed into a two dimensional plane using the homography matrix and used to compute player distance and speed based on euclidean distance.
The sistem is evaluated on test clips from the SoccerNet-Tracking test set. Based on the HOTA metric, the combination of ByteTrack with GTA demonstrates better tracking performance compared to ByteTrack alone. However, the projected player positions remains unstable, resulting in limited accuracy in performance evaluation. This highlights the need for improvements in the projection stage to enable more precise and consistent player performance analysis.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Pelacakan Multi Objek, Performa Pemain, Sepak Bola, YOLOv11, Multi Object Tracking, Player Performance, Soccer, YOLOv11
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Syuhada Ghifari Maaliki Syafa
Date Deposited:	29 Jul 2025 07:32
Last Modified:	29 Jul 2025 07:54
URI:	http://repository.its.ac.id/id/eprint/122820

Actions (login required)

View Item