Localisation of domain specific objects in esports video streams with partially random layouts

Important results from the project

The aim of the project was to improve a partly manual process for generating data using "computer vision" (CV). This is to enable the technology to be applied to a wider range of tournaments and matches and to extract more data points with higher precision. The goals together aim to make the computer vision based products more competitive. Furthermore, the company´s goal was to learn more about project management and development work within ML/AI in order to improve the application over time and work efficiently with similar projects.

Expected long term effects

The company regards the project as very successful. The new system was launched during the project period and was validated as achieving a very high degree of automation and that the number of covered tournaments and matches increased significantly, about 7 times as many. With a larger amount of data points generated. On the market side, the effects have been positive and customers with higher requirements have been able to be approached. The company has also started to implement the system on more games successfully and sees more opportunities for expansion in ML/AI in the future.

Approach and implementation

To achieve as high a standard as possible in terms of both the technical solutions and process/project management, the company chose to turn to experts in the field at the onset of the project. The experts worked together with and guided the company through the first half of the project. Together, different approaches to the problem were tested, after which demos for the most promising methods were produced. Finally, the company developed a final solution on its own. This approach proved effective in achieving hard results and to internalize the experts´ knowledge.