Automatic Extraction of Heat Maps and Goal Instances of a Basketball Game Using Video Processing

The use of image processing techniques and computer vision to obtain teams statistics in different sports, currently represents a new source of information very useful for season preparedness. In this work, we propose different methods to show the players position for a specific time, to perform the point counting, and to extract clips of goal situations in the match. In order to accomplish this, we use a combination of transfer learning using a pre-trained deep neural network with a database of basketball game excerpts, and video processing techniques. As a proof of concept, the method was applied to a basketball game of local teams, showing the feasibility of the proposed approach.