Our Technology

Mouse DimensionsiBehaveŽ is the core technology at the heart of Actual's tailored behaviour analysis solutions and uses a unique three step, approach to identify behaviors as defined by the researcher, and not the vendor.

Learning:

Given a collection of representative sample videos in which the diversity and range of a particular behaviour is evident, iBehaveŽ can "learn" to recognize these in a semi-supervised manner. That is: the user simply annotates the training videos when the behaviour occurs, and the system will then discover the patterns of motion and context that define that behaviour driven from the data.

This unique signature can then be used by iBehaveŽ to identify similar patterns in previously unseen videos. Capturing in essence the researchers interpretation of that particular behaviour.

Typically we require only 1-2 hours of video footage to reliably identify behaviors.

Validating:

Video sections which have not yet been included in the analysis are used as a blind test for the system. During this phase the accuracy is measured. Any inconsistencies in the sample video will be discussed with the customer and resolved.

Identifying:

Following the generation of a signature, iBehaveŽ uses the associated rules to identify behaviors in an unlimited volume of additional video footage. This ability to scale completely removes the requirement for human raters but still gives the confidence that identified behaviors will match those required for this trial. iBehaveŽ initially identifies the subject's movements from the video footage then uses this data against the rules to indicate when the required behavior is in progress.

An unlimited number of subsequent footage can then be analyzed by the system and results output in Microsoft Excel™ format or integrated with your existing data analysis systems.

Publications

The academic research behind iBehaveŽ has been published in a number of peer-reviewed international conferences and workshops:

  • T.C. Lukins, J.M. Young, R.R. Sillito, and J.D. Armstrong. Fly, Camera, Action! Techniques to Improve the Video Capture and Storage for Automated Analysis of Drosophila Behavior Experiments. Under Review. 2010.
  • J.A. Heward, M.A. Dewar, E.C. Wright, and J.D. Armstrong. Who watches the behavior experts? Under Review. 2010.
  • R.R. Sillito, T.C. Lukins, and J.D. Armstrong. Improved rodent contour extraction using a priori shape information. In Workshop on the Visual Observation and Analysis of Animal and Insect Behavior (held at ICPR'10), Istanbul, Turkey, 2010.
  • M.A. Dewar, T.C. Lukins, J.A. Heward, and J.D. Armstrong. Classification of animal behaviour using dynamic models of movement. In Workshop on Stochastic Models of Behaviour (held at NIPS'08), Whistler, Canada, 2008.
  • T.C. Lukins, M.A. Dewar, J.A. Heward, and J.D. Armstrong. Extracting drosophila courtship behaviour statistics from video. In Workshop on the Visual Observation and Analysis of Animal and Insect Behavior (held at ICPR'08), Tampa, USA, 2008.
  • P.A. Crook, T. C. Lukins, J.A. Heward, and J.D. Armstrong. Identifying semi-invariant features on mouse contours. In Proceedings of the British Machine Vision Conference, Leeds, UK, 2008.
  • T.C. Lukins, M.A. Dewar, P.A. Crook, J.A. Heward, A.B. Hawcock, and J.D. Armstrong. Automatically determining active investigation in rodents using contour analysis. In Proceedings of the 6th Measuring Behaviour Conference, Wageningen, NL, 2008.
  • J.A. Heward, P.A. Crook, T.C. Lukins, and J.D. Armstrong. iBehave - application of supervised machine learning to behaviour analysis. In Proceedings of the 6th Measuring Behaviour Conference, Wageningen, NL, 2008.
  • J.D. Armstrong, D.A. Baker, J.A. Heward, and T.C. Lukins. Sex, flies and no videotape. In Proceedings of the 5th Measuring Behavior Conference, Wageningen, NL, 2005
  • J.A. Heward, D.A. Baker, T.C. Lukins, and J.D. Armstrong. flyTracker: real-time analysis of insect courtship. In Proceedings of the 5th Measuring Behavior Conference, Wageningen, NL, 2005