The main task of the sensor fusion software is to analyse the recorded videos including their audio tracks and to generate match statistics. The following statistics can be generated or have been considered in the architecture to be implemented in the future:
In order to be able to detect goals and outs, it is necessary to recognise where the ball is in relation to the field lines. To do this, an algorithm is used that detects these lines. The result of this algorithm is then filtered and interpolated so that the result can be used to detect events.
Team recognition is used to assign the detected events (goal, free kick, foul, etc.) to the corresponding teams. Estimated values of the teams' jersey colours are specified beforehand via the configuration interface. The detected players are cut out, their colour environment analysed and compared against the configured values.
The whistle recognition uses the Fast Fourier Transform algorithm, which divides the long-term signal into several shorter signals of equal length, and then calculates the Fourier Transform of each shorter segment separately. Taking the maximum Amplitude in a specific frequency band(3kHz-5kHz), which is the spectrum of the whistle sound.
After the match analysis, the results are saved in a file in JSON format so that they can be visualised in an independent GUI. This makes it easy to track the timing of match events and the general course of the match. Optionally generated videos can be displayed in the GUI at the same time as the match results.
For the analysis of different games, certain parameters have to be adjusted (e.g. approximate colours of the players' jerseys in BGR format, log level, etc.). These can be set via a command interpreter before starting the actual analysis.