|Ph.D Student||Zohar Ron|
|Subject||Flow Conservation Group Tracking|
|Department||Department of Computer Science||Supervisor||Professor Dan Geiger|
|Full Thesis text|
This research develops means needed for robust group tracking and group identification systems. Group tracking is an approach to tracking where groups of objects are being monitored as a single entity rather than monitoring each object individually. Group tracking is applied to tracking closely spaced objects with similar state vectors. For such closely spaced objects, it is often impossible to track each object separately due to sensors' limitations.
In this thesis we describe a novel framework for group tracking and identification termed Flow Conservation Group Tracking, which we believe to be a preferred extension of the methodology from tracking single objects to tracking groups of objects. Our framework integrates local kinematic measurements with flow conservation constraints so as to yield a more robust estimation of the position and types of objects across an arena. We base our framework to a large extent on the ability to estimate and match the number of targets in each group, in addition to kinematics.
The Flow Conserving Estimation of Number of Objects component of our framework focuses on the problem of accurately estimating the number of targets in an arena. When a group in the arena splits into several subgroups, it is clear that the number of targets in the subgroups equals the number of targets in the original group. Consequently, it is possible, given a set of measurements of the number of targets on each track, to globally estimate the most probable number of vehicles on each track by utilizing these flow conservation constraints. We model this problem using the notion of flow networks.
Our Flow Conserving Type Identification component is applicable to systems which receive aggregated type information about groups in the arena. We develop a framework and polynomial algorithms for the fusion of reports in the context of global estimation in flow networks. Instead of providing a summary report that is solely based on measurements received on a single edge, we address the problem of generating a summary report per edge in a global fashion that uses reports on other edges in the flow network to correct its estimates.