Faculty of Informatics Vienna University of Technology Institute of Computer Aided Automation PRIP Home PRIP Home
Personal tools
You are here: Home Teaching PRIP-Talks Andreas Zweng: Unexpected Human Behavior Recognition in Image Sequences

Andreas Zweng: Unexpected Human Behavior Recognition in Image Sequences

Spezifikationsvortrag Diplomarbeit

  • Presentation
When Nov 24, 2009
from 04:35 pm to 04:55 pm
Where Sem 183/2
Add event to calendar vCal
Surveillance applications (i.e Event Recognition) typically use Tracking to recognize abnormal behavior. Therefore it is necessary to define these events by a sequence of positions of persons, the trajectories. The trajectories of human movement can also be trained to create a more general system, but since tracking requires human recognition which is done with the help of motion detection and background modelling algorithms it requires robust results from these preprocessing steps. Event Recognition in turn requires robust results from the tracking algorithm to obtain an applicable output. Experimental observations showed, that people perceive unexpected behavior by abstract features like the crowd density, person positions or local activity. They tend to the extremes while judging the probability of abnormality in image sequences meaning that an image in a sequence is either judged as completely unexpected or completely normal. The only uncertainty is in the transition of normal and unexpected behavior whereas this uncertainty appears in only 3-10 frames depending on the framerate of the image sequence and is continuously decreasing (the certainty that the actual image is abnormal is increasing). This thesis covers training and classification algorithms for the discussed featuresand aims for general methods in behavior recognition where the values of the processed features are treated seperately before combining them to compute a probability of unexpectedness for each frame in the image sequence.
Document Actions