Multitarget visual tracking based effective surveillance with cooperation of multiple active cameras.

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Effective surveillance requires tracking multiple moving objects simultaneously.
• Coordinating multiple cameras presents challenges in maintaining coverage and maximizing tracked targets.

Purpose of the Study:

• To develop a robust, multi-camera surveillance system for real-time tracking of multiple objects.
• To enhance target tracking accuracy, even with overlapping objects.
• To optimize camera selection and task assignment for wide-area monitoring.

Main Methods:

• Developed distributed camera agents for independent multi-object tracking.
• Extended particle filters with target depth estimation for handling overlapping targets.
• Proposed a mutual information and Monte Carlo-based strategy for suboptimal camera action selection.
• Introduced a hierarchical online position strategy for camera selection and task assignment.

Main Results:

• Achieved near real-time response in tracking multiple moving objects.
• Demonstrated effective tracking of overlapping targets using enhanced particle filters.
• Successfully maximized the number of tracked targets within a wide surveillance area.
• Verified system performance through computer simulations and extensive experiments.

Conclusions:

• The proposed multi-camera surveillance system effectively tracks multiple moving objects in near real-time.
• The integrated camera coordination strategies ensure efficient coverage and maximize target tracking.
• The system offers a scalable solution for wide-area surveillance applications.