Systems and methods for counting repetitive activity in audio video content
View Patent ↗Repetitive activities can be captured in audio video content. The AV content can be processed in order to predict the number of repetitive activities present in the AV content. The accuracy of the predicted number may be improved, especially for AV content with challenging conditions, by basing the predictions on both the audio and video portions of the AV content.
1. A method for predicting a number of repetitions in an audio/video (AV) content comprising:
extracting video features from a video portion of the AV content;
extracting audio features from an audio portion of the AV content;
processing at least a portion of the extracted video features and a portion of the extracted audio features to determine a temporal stride for use in processing the video portion of the AV content;
processing the extracted video features with a neural network to predict a number of repetitive actions in the video portion based on the determined temporal stride;
processing the extracted audio features with a neural network to predict a number of repetitive actions in the audio portion
predicting a final number of repetitive actions from the predictions of the number of repetitive actions in the video portion and audio portion and the extracted video and audio features.
2. The method of claim 1 , wherein the video portion of the AV content comprises a video clip from the AV content, and the audio portion of the AV content comprises a spectrogram of the AV content.
3. The method of claim 1 , wherein the audio features are extracted using a 2D residual neural network.
4. The method of claim 3 , wherein the residual neural network comprises at least 18 layers.
5. The method of claim 1 , wherein the audio features are extracted using a 3D residual neural network.
6. The method of claim 1 , wherein the neural network used to process the extracted video features comprises at least one fully connected layer.
7. The method of claim 6 , wherein the neural network used to process the extracted video features comprises:
a first fully connected layer outputting a repetition count of each of a plurality of repetition classes present in the video portion of the AV content; and
a second fully connected layer classifying the plurality of repetition classes present in the video portion of the AV content.
8. The method of claim 1 , wherein the neural network used to process the extracted audio features comprises at least one fully connected layer.
9. The method of claim 6 , wherein the neural network used to process the extracted audio features comprises:
a first fully connected layer outputting a repetition count of each of a plurality of repetition classes present in the audio portion of the AV content; and
a second fully connected layer classifying the plurality of repetition classes present in the audio portion of the AV content.
10. The method of claim 9 , wherein predicting the final number of repetitive actions comprises a residual block processing extracted audio features and a fully connected layer processing features output from the residual block and video features.
11. A non-transitory computer readable memory storing instructions, which when executed by a processor of a system configure the system to perform a method comprising:
extracting video features from a video portion of the AV content;
extracting audio features from an audio portion of the AV content;
processing at least a portion of the extracted video features and a portion of the extracted audio features to determine a temporal stride for use in processing the video portion of the AV content;
processing the extracted video features with a neural network to predict a number of repetitive actions in the video portion based on the determined temporal stride;
processing the extracted audio features with a neural network to predict a number of repetitive actions in the audio portion
predicting a final number of repetitive actions from the predictions of the number of repetitive actions in the video portion and audio portion and the extracted video and audio features.
12. The non-transitory computer readable memory of claim 11 , wherein the video portion of the AV content comprises a video clip from the AV content, and the audio portion of the AV content comprises a spectrogram of the AV content.
13. The non-transitory computer readable memory of claim 11 , wherein the audio features are extracted using a 2D residual neural network.
14. The non-transitory computer readable memory of claim 13 , wherein the residual neural network comprises at least 18 layers.
15. The non-transitory computer readable memory of claim 11 , wherein the audio features are extracted using a 3D residual neural network.
16. The non-transitory computer readable memory of claim 11 , wherein the neural network used to process the extracted video features comprises at least one fully connected layer.
17. The non-transitory computer readable memory of claim 16 , wherein the neural network used to process the extracted video features comprises:
a first fully connected layer outputting a repetition count of each of a plurality of repetition classes present in the video portion of the AV content; and
a second fully connected layer classifying the plurality of repetition classes present in the video portion of the AV content.
18. The non-transitory computer readable memory of claim 11 , wherein the neural network used to process the extracted audio features comprises at least one fully connected layer.
19. The non-transitory computer readable memory of claim 16 , wherein the neural network used to process the extracted audio features comprises:
a first fully connected layer outputting a repetition count of each of a plurality of repetition classes present in the audio portion of the AV content; and
a second fully connected layer classifying the plurality of repetition classes present in the audio portion of the AV content.
20. The non-transitory computer readable memory of claim 19 , wherein predicting the final number of repetitive actions comprises a residual block processing extracted audio features and a fully connected layer processing features output from the residual block and video features.
21. A system comprising:
a processor for executing instructions; and
a memory storing instructions, which when executed by the processor configure the system to perform a method comprising:
extracting video features from a video portion of the AV content;
extracting audio features from an audio portion of the AV content;
processing at least a portion of the extracted video features and a portion of the extracted audio features to determine a temporal stride for use in processing the video portion of the AV content;
processing the extracted video features with a neural network to predict a number of repetitive actions in the video portion based on the determined temporal stride;
processing the extracted audio features with a neural network to predict a number of repetitive actions in the audio portion
predicting a final number of repetitive actions from the predictions of the number of repetitive actions in the video portion and audio portion and the extracted video and audio features.