Automatic Pass Annotation from Soccer Video Streams Based on Object Detection and LSTM

Soccer analytics is attracting increasing interest in academia and industry, thanks to the availability of data that describe all spatio-temporal events occur each match. These (e.g., passes, shots, fouls) are collected by human operators manually, constituting a considerable cost for providers terms time economic resources. In this paper, we PassNet, method recognize most frequent soccer, i.e., from video streams. Our model combines set artificial neural networks perform feature extraction streams, object detection identify positions ball players, classification frame sequences as passes or not passes. We test PassNet on different scenarios, depending similarity conditions match used training. results show good significant improvement accuracy pass with respect baseline classifiers, even when match’s training sets considerably different. first step towards an automated event annotation system may break costs annotation, enabling collections minor non-professional divisions, youth leagues and, general, competitions whose matches currently annotated providers.