查看文章

One of the common faults in volleyball is double contact while setting the ball for a spike. It is hard to detect this fault by the players. Even the referees sometimes find it difficult to observe. In this work, we propose an automatic double contact fault detection approach using a single camera in outdoor volleyball video. The video is first analyzed to detect and track the ball; the bounding boxes are then processed to extract a deep Spatio-temporal representation using a state-of-the-art 3D-convolution-based neural network, which is finally fed to a multilayer perceptron for classification. To the best of our knowledge, this is the first work on volleyball double-contact detection. The proposed framework achieves an average accuracy of 77.16% on 5-fold cross-validation. The framework is useful for players during training and for referees as a decision-support tool.