The great competitive results of Slovenian ski jumpers in world cup and continental competitions have sparked much interest with regard to more active participation in this attractive sport. At junior levels it is now normal for national competitions to have more than 100 jumpers. However, expensive and logistically demanding commercial video distance measuring tools are only used at the top-level competitions (world cup, continental cup). We are experimenting with two affordable approaches to automate and speed up video distance measuring. The first approach uses a deep convolutional neural network with 10 hidden layers in order to automatically detect the correct landing frame. Each frame is classified either as in the air' or `on the ground'. This approach achieves very high classification accuracy for determining the type of frame. However, as errors always occur near the correct landing frame, human intervention is still necessary. The second approach utilises classic computer vision image segmentation techniques to acquire the positions of a ski jumpers skis and legs in order to determine the correct landing point within the frame, and therefore the distance based on the measuring grid (currently with an accuracy of 0.5-1 m).

We are optimistic that the system will be effective for practical use on small hills in the near future. However, moving to larger hills will require additional research in order to allow for two, three or four network cameras. The system still needs further testing (especially the automated components) under artificial lighting conditions (for night competitions).