Crowdsourcing Annotations for Video Retrieval in Sports Videos (Bachelor Thesis, Finished)


Fabio Sulser


The project will enable sport savvy users (not necessarily sports scientists) in a crowdsourcing environment to view video snippets of sport games (soccer/football) and annotate an array of information about each video snippet they see. The position of the ball, the players involved and their position, the event(s) taking place will be annotated. Several events can be annotated and linked to a single scene. Multiple users can view the same scene and if the majority agree to the event, and pick a location that is in proximity to the other users, the data is validated and saved. The interface should be rather fun and simple to the end users, however, they must be sport savvy and know the rules of the game (soccer/football) . On one side of the screen, there will be a video player that will play a video snippet from a soccer game. The user will view the scene and can replay it as often as he or she wants. On the opposite side of the screen, there will be an empty soccer field. It will be used to place the location of the ball and players involved in the video scene. An array of check boxes containing potential events will be displayed and the user can check one or more of the events depending on the scene. The timing of the events, in relation to the game, will be synchronized and saved automatically without the need for user input.

The system will be developed on one of the crowdsourcing platform such as Micro Workers or Amazon Mechanical Turk and its output will be used within the SportSense project being developed in the department.

Start / End Dates

2014/02/06 - 2014/06/05


Research Topics