Identifying Interesting Moments in Controllers Work Video via Dimensionality Reduction

We explore use of machine learning in automating the discovery of meaningful time intervals in video data. We combine Convolutional Neural Networks and Principal Component Analysis in order to zoom-in on interesting moments in hours-long videos of air traffic controllers work. Experimental results for air traffic control tower at Stockholm Bromma airport confirm feasibility of our approach. The method may be consequently used to single out workload-influencing factors, incident investigation and other post-operational analysis of controllers performance.