Object-centered Fourier Motion Estimation and Segment-Transformation Prediction

Moritz Wolter, Angela Yao und Sven Behnke
In proceedings of 28th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, 2020
 

Abstract

e ability to anticipate the future is essential for action planning in autonomous systems. To this end, learning video pre-diction methods have been developed, but current systems often pro-duce blurred predictions. We address this issue by introducing an object-centered movement estimation, frame prediction, and correction frame-work using frequency-domain approaches. We transform single objects based on estimated translation and rotation speeds which we correct us-ing a learned encoding of the past. This results in clear predictions with few parameters. Experimental evaluation shows that our approach is accurate and efficient.

Stichwörter: Fourier, Prediction

Source code available at: https://github.com/v0lta/Fourier-Motion-Estimation-and-Segment-Transformation

Bilder

Bibtex

@INPROCEEDINGS{wolter2020object,
     author = {Wolter, Moritz and Yao, Angela and Behnke, Sven},
      title = {Object-centered Fourier Motion Estimation and Segment-Transformation Prediction},
  booktitle = {28th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine
               Learning},
       year = {2020},
   keywords = {Fourier, Prediction},
   abstract = {e ability to anticipate the future is essential for action planning in autonomous systems. To this
               end, learning video pre-diction methods have been developed, but current systems often pro-duce
               blurred predictions. We address this issue by introducing an object-centered movement estimation,
               frame prediction, and correction frame-work using frequency-domain approaches. We transform single
               objects based on estimated translation and rotation speeds which we correct us-ing a learned
               encoding of the past. This results in clear predictions with few parameters. Experimental evaluation
               shows that our approach is accurate and efficient.},
        url = {https://vi.informatik.uni-bonn.de/papers/ESANN_2020_Wolter.pdf}
}