Samenvatting

The graduation project of Andrii Roizin, a software engineering student at Saxion, was developed as a part of the Sportkeur 2.0 project. This project is led by Go2Sure and the Saxion Ambient Intelligence research group is one of its partners.
            The purpose of the project was to discover the possibility of automating the process of inspecting possible hazardous situations of the arenas/tracks used for biking events using computer vision. This checkup is performed by a safety inspector. The idea was to develop such a solution with the help of semantically segmented data to detect hazardous situations like poor road conditions, unclear track directions, steep turns, missing signboards, etc using the videos from the track.
            The research and development resulted in a software system that is able to take the video from the biking track as an input, analyze it using computer vision techniques, and output the semantically segmented video. The problem was solved using modern computer vision and software technologies and the solution satisfies the requirements of Go2Sure and the AMI research group. The semantically segmented videos were created using the DeepLabV3+ machine learning architecture. It reached 76.2% of mIOU on the validation data and was able to deal with the traffic situations on Dutch roads. Likewise, the machine learning solution was integrated into the existing SportKeur system using Django API. At the same time, there is room for future improvements and the possibilities for the continuation of the project are described in the document alongside the explanations of the important decisions.