Conventional tracking
approaches assume proximity in space, time and appearance of objects in
successive observations. However, observations of objects are often widely
separated in time and space when viewed from multiple non-overlapping
cameras. To address this problem, we present a novel approach for establishing
object correspondence in a system of non-overlapping cameras. We observe
that people or vehicles follow the same paths in most cases, i.e roads,
walk ways, corridors etc. Our method exploits this redundancy in paths
traversed by using both motion trends and appearance of objects for tracking.
Our system does not require any inter-camera calibration, instead the
system learns the camera topology and path probabilities of objects using
Parzen windows during a training phase. Once the training is complete,
correspondences are assigned using the maximum a posteriori (MAP) estimation
framework. The learned parameters are updated with changing trajectory
patterns. Experiments with real world videos are reported which validate
the proposed approach.
Omar Javed, Khurram Shafique, Zeeshan Rasheed and Mubarak Shah
Omar Javed, Khurram Shafique, Zeeshan Rasheed and Mubarak Shah For Power Point presentation click here
|