Fig 1.
General diagram of the proposed method.
Fig 2.
Four consecutive frames from a fight sequence where the four largest motion blobs have been marked.
Fig 3.
Four consecutive frames from of a non-fight sequence where the four largest motion blobs have been marked. Compare with those of Fig 2. In the previous figure the motion blobs are larger and clustered, whereas in the current figure the motion blobs are smaller and not clustered.
Fig 4.
Four consecutive frames from with global movement where the four largest motion blobs have been marked.
Fig 5.
Sample fight videos from the action movie (top) dataset and the Hockey (bottom) datase
Fig 6.
The 101 actions in UCF101 shown with one sample frame.
Fig 7.
Sample frames of Sumo and Punch categories in UCF101.
Table 1.
This table shows the mean and standard deviation of the Accuracy for these three classifiers for variant 1 of the proposed method.
Table 2.
This table shows the mean and standard deviation of the Accuracy for three classifiers for variant 2 of the proposed method.
Table 3.
This table shows the results for the three datasets using Support Vector Machines (SVM), AdaBoost and Random Forests (RF) classifiers for the related methods and the two proposed variants.
Three measures have been calculated: mean accuracy, standard deviation accuracy and AUC.
Fig 8.
ROC curves for the five related methods and the two considered variants. The Random Forests classifier on Movies dataset is used.
Fig 9.
ROC curves for the five related methods and the two considered variants. The Random Forests classifier on Hockey dataset is used.
Fig 10.
ROC curves for the five related methods and the two considered variants. The Random Forests classifier on UCF101 dataset is used.
Fig 11.
ViF results comparing accuracy vs feature extraction time.
Fig 12.
LMP results comparing accuracy vs feature extraction time.
Table 4.
Feature extraction times. Average times measured with the non-fight videos in the UCF101 dataset, on an Intel Xeon computer with 2 processors at 2.90Ghz.
Table 5.
Sub-steps of the proposed method.