Real-time automated concurrent visual tracking of many animals and subsequent behavioural compilation

Abstract

One of our major research focus areas is real-time visual tracking and monitoring of moving and static objects in a video sequence. In particular we are interested in (1) object localization (also referred to as the focus of attention) which involves identifying the object of interest, (2) tracking the object using a model of the object which was initiated in step 1, and (3) understanding the accumulation of movements of the object over time (i.e., behavior). The objects of interiest for the purpose of this proposal are pigs. Automatically monitoring pigs via a non-invasively placed camera in their pens is interesting because the pigs are monitored in their natural habitat. Visual tracking involves modelling the object of interest and keeping track of its position and orientation through time. Issues include tracker recovery from error and preventing the tracker from jumping to other pigs. we have been able to demonstrate tracking pigs at about 10-15 Hz, however, the tracker tends to drift off the target eventually. We have only experimented with a single pig but our initial tests indicate that we can probably track at least 10 pigs simultaneously. Some unknowns include determining how quickly the pigs move and the type of motions, including quick jerky movements. Our preliminary investigations revealed that a blob tracker is insufficient for producing accurate traces.