The idea was initially inspired to bring people closer to virtual representations of players that match characteristics they feel they possess. In this way people can engage and get to know players they may not have any idea before by providing their estimated attributes. The second reason for this project is for scouting players. The advantage of VR allows for a close to real representation of physical characteristics of each player for instance (height and weight), currently we change the height of the players and a user with a head mounted display could have a better idea of how he compares in height to Messi or Ronaldo for instance.

What it does

The interface allows for input feature values which are received by our algorithm. The algorithm searches the feature space for the 10 Nearest Neighbors and returns a list of those players. We then generate a visual representation of the player changing a physical characteristic like height. We position the player in the environment in which a user eventually with a VR headset can compare to. Information of the player is added to the 3D interface and can be compared side by side with the values input by the user.

How we built it

A set of the most representative features were initially selected by using PCA. We decided to reduce the number of features to give the user a less tedious experience (and to reduce multicollinearity). All features were scaled to have a mean of 0 and a standard deviation of 1. Then using the Euclidean distance, we found the 10 nearest neighbors in the feature space. The application follows a Client - Server approach. The server side receives the feature values to match from players and returns to the client in JSON format a list of players. The information is parsed on the client and presented to the user on an immersive way.

Challenges we ran into

We did not have appropriate hardware so we had a hard time using our phones as VR displays. We just had an android 6.0 with some limitations at deployment time. We are looking into a mobile experience and a Gear VR would be the best approach. In addition, we created a prediction model using linear regression in order to predict the Transfer Market Value for each individual player; however, we faced issues due to the inconsistency of names while merging the data from the FIFA 18 and Transfer Market datasets.

Accomplishments that we're proud of

We successfully divided the tasks and could integrate different parts in a seamless way. We are happy that we complete most of our initial milestones.

What we learned

Teamwork, collaboration, testing, data analysis

What's next for SoccerMatchVR

We would like to generate virtual behaviors based on each player individual characteristics. For instance, calculate probabilities of a successful tackle or shot and visualize it in a VR environment. Additionally, we would like to improve our models to provide a closer to real life representation of the players. VR can be used for analysis of match trends and for a possible better understanding of each other teams strategies.

Share this project: