Unusual motion detection


Unusual motion: detection of anomalous events based on general motion information.
  • Detection of anomalous events in urban traffic scenes. A longer term training phase (online/offline) is necessary to learn the typical/usual motion patterns. In the later detection phase motion irregularities will be signaled w.r.t. learnt motion areas.
  • The unusual motion detector module can be used in situations where:
    • there is a static camera
    • in daylight
    • at ~8-10m height (or higher)
    • overlooking a region where there is general motion, and we expect to have unusual movement occur sometime in the future
    • detection is preceded by a training period: offline (before running) or online (when the module starts, learns for a specific period, then starts the detection phase)
  • Method:
    • extract foreground + optical flow --> build statistics, classify motion directions
    • signal differences w.r.t. learnt motion direction statistics
  • Availability:
    • For Windows or Linux
    • Library versions (for C++ and for Java/JNI)
    • Realtime performance
  • Connected publications:
    • Á. Utasi, "Novel Probabilistic Methods for Visual Surveillance Applications", Dissertation, Doctoral School of Information Science and Technology, University of Pannonia, Veszprém, Hungary, January 2012
    • Á. Utasi, L. Czúni, "Anomaly Detection with Low-Level Processes in Videos", in Proc. of The 3rd International Conference on Computer Vision Theory and Applications, pp. 678-681, Funchal, Portugal, 22-25 January 2008

In the video regions with red color show unusual motions.

Contact: Ákos Utasi