SMART FOOTBALL: IMPLEMENTATION OF PLAYER TRACKING WITH PLAYER LOCALIZATION ON FOOTBALL MATCH VIDEOS

• Main Author: Stefen Mardianto, Thomas
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/86188
• Institution: Institut Teknologi Bandung

As a popular sport, soccer game can be improved in quality through technological innovations in the analysis of match videos. Football match videos in Indonesia are available and easily accessible through broadcasts both television and streaming applications. However, there is no technology that is able to track the player's movements from a video, which will be useful for game strategy analysis. There is a challenge in implementing this technology, namely the characteristics of broadcast videos that have a change in camera perspective. This challenge can be solved using deep learning models and computer vision to detect the position of the field and calibrate the camera. With the help of the player object detection model that has been developed, the player localization feature can be created.This feature is able to track the movement of players relative to the field to overcome the challenges of broadcast video characteristics and generate data on the results of player tracking on the field. The methods used include semantic segmentation of field elements that results in the detection of keypoints and field lines and camera calibration that produces homography to transform the coordinates of objects from imagery to the real world. The results of the study show that the player localization feature developed is able to achieve quite high accuracy with a final score of 73.7 and an IoU of 98.6%. These findings show great potential in video analysis of football matches through the application of player localization to player tracking to generate insights in evaluating game strategies.