LiveCalcio

Abstract

LiveCalcio is a cross-platform mobile application for both Android and iOS devices. It’s a revolutionary mobile application that allows football enthusiasts to upload a football video of a Manchester United match from the 2023/24 season onto the application, and then tap on any Manchester United player on the field and get their match statistics. The user can view the minutes played, match rating, goals scored, assists, tackles, shots, passes, fouls, duels won, offsides and fouls drawn of the player they tapped on. The application works by uploading the video to the Firebase Store upon a tap, and then calculating the x,y coordinates of the tapped area and mapping them to the original video. The request is sent to the model pipeline, which is deployed as an API through FastAPI on Google Colab using ngrok. The model first establishes a few checks to determine whether the area tapped on is a person (as opposed to the field, ball, crowd, or anything of that nature), and whether the tapped-on person is a Manchester United player. It does this by using an HSV Classifier, which checks whether the kit colour of the tapped-on player is within the red colour range of the Manchester United kit. After the initial checks have passed, the video goes through two phases of the model pipeline. The first phase identifies the tapped-on person’s bounding box using an out of the-box YOLOv8 model. The second phase uses a YOLOv5 model pre-trained on a jersey number dataset, and uniquely identifies the person through their jersey number on each frame that the player appears. Majority voting is done on the results of the jersey number classifications and is sent as a response to the mobile application. The application then displays the statistics of the Manchester United player in their previous game (or current game if one is currently on-going) using the jersey number and the statistics API. The user is also able to view the progression of each statistic against the historical record achieved by a Manchester United player in that statistic, given they have the appropriate subscription tier. The historical record of each statistic is sourced from StatsMuse and is stored within a Firebase Datastore collection. Additionally, the user is also able to register, login, and view their profile details through separate screens.