deletions | additions       diff --git a/section_Introduction_The_current_works__.tex b/section_Introduction_The_current_works__.tex  index 038abc7..eccaf30 100644  --- a/section_Introduction_The_current_works__.tex  +++ b/section_Introduction_The_current_works__.tex 
...
The current works on soccer match analysis used extra high level data and annotation (such as whoscored, transfertmark) which require human annotation and so preprocessing. In this project we limited our input to the video of a match filmed with a multi-camera system. We focus on an unsupervised classification of the players based on clustering of their trajectories which are extracted automatically from the video of a match during a short period of time. This information is deeply woven into the fabric of team strategy analysis since it is related to the global placing of the players and the centers of the cluster may correspond to the leader of the game at that period. Thus we managed to find patterns in the paths of the players clustering players trajectories.   The problem of tracking players with multiple camera have been achieved consistently for instance in \cite{Ben_Shitrit_2011} even if their paths may intersect over long period of time.   On another hand the clustering process is closely linked to the choice of a metric between the object we want to classify: here the players' trajectories. Most of the time the metrics used are based on euclidian metric. However only discrete trajectories are available in the form of array which can be of different sizes with different time discretization or with different speed and  euclidian metric are irrelevant to compare trajectories trajectories,  mainly because it doesn't take the time parameter into account. So, in our case, a relevant clustering process should be decomposed into 2 parts: \begin{itemize}  \item Compute the distances matrix between the trajectories   \item Use a general clustering algorithm which is independent of the metric used  \end{itemize}  Hence we used the clustering algorithm described in [?] which take a distance matrix as single input. Furthermore it automatically selects the clusters' centers and the number of clusters.  Our contribution consist of a relevant similarity measure between trajectories based on Mutual Information which allows the clustering of the players' trajectories according to the interdependency of their path.  Yet only discrete trajectories are available in the form of array which can be of different sizes with different time discretization or with different speed.