Motion Computation Without Iterative WARPing

Peleg Shmuel, HUJI, School of Computer Science and Engineering, Computer Science
Rav-Acha Alex, HUJI, School of Computer Science and Engineering

Fast and simple editing of special video effects 

  • Motion computation in video editing involves warping or manipulations of time in order to control the chronological time of events, including delaying or slowing down some activities while advancing or speeding up others.
  • The dominant method for motion computation is the iterative motion analysis proposed by Lucas& Kanade (LK), used for both parametric motion computations and object tracking. The method requires substantial computational cost.
  • Our method avoids the iterative image warping used in the original method by using a single iteration.
  • The multiframe alignment overcomes the high complexity or restrictive assumptions (such as small motion or large memory) drawbacks of current multiframe alignment methods
  • Experimental results show improvement in both complexity and accuracy

Our Innovation

  • Implements LK with a computational cost of a single iteration without affecting the accuracy of the estimated motions. Combined with a multiframe alignment to obtain a fast and robust alignment.

Key Features

  • Computer vision software for the extraction and analysis of information from images
  • Appropriate to online and real time applications

The Opportunity

  • Motion computation is used for applications in: medical remote examination, security visual surveillance, robotics, intelligent transportation, mixing real and virtual objects (AR/VR), moving object tracing and video analytics  

Patent Status

Granted US 7,894,528

Contact for more information:

Anna Pellivert
Contact ME: