a b s

This project seeks to investigate multi-target tracking through network camera coordination.  The issues involved with this model are resolved by first: having reliable detection and prediction algorithms for tracking non rigid objects and second: coordinating the camera network with known camera positions and orientations. Object retention was accomplished through the color based particle filter which sorts a sampled image into a color histogram and predicts the position in the next frame through a dynamic stochastic model. In order to achieve accurate coordination, the cameras were mapped according to a central reference point based on Jacobian transformations, allowing the cameras to share a vision network.  The combination of these two methods provided reliable performance even under complex environments where the cameras were unable to see one another.


Group Members : Robert Amos, Aldo Gutierrez, Daniel Kruse.

Advisor : Dr. Yuichi Motai.