Output details
15 - General Engineering
Heriot-Watt University (joint submission with University of Edinburgh)
Tracking With a Hierarchical Partitioned Particle Filter and Movement Modelling
Although human tracking from video sequences is ubiquitous, this work is significant for two reasons. First it performed full articulated-tracking of the human body unlike the much more common tracking of fixed (or even deformable) templates. Second, here the tracked-torso and limb-positions were coupled with a behavioral model (output-4) that demonstrated action recognition. The software was fully evaluated by independent competition (http://vision.cs.brown.edu/humaneva/index.html), which showed joint tracking errors of the order of 10cm in complex, unseen video footage. This work has informed the software development for the current EU Locobot project (FP7 260101, €3.7M).