Coopetitive Multi-Camera Surveillance using Model Predictive Control
Basic Idea Conceptual Motivation Key Contributions

Practical Implementation
Dual Camera setup

Practical Implementation
Triple Camera Setup
People Publications

Practical Implementation/Experimentation

a) Dual camera setup

Let us look at the scenario of an ATM lobby or a museum subsection with single entry point, one important artifact and 2 cameras[Figure 1]. Our aim is to contiuously obtain frontal facial images of the intruder entering the ATM lobby even if he intentionally tries to prevent such face capturing. Our experiments showed that Coopetitive interaction approach performed better than other possible alternatives.

Video demos for one simple trajectory undertaken by intruder have been presented here.

Approach Comments Demo Video
Only Competition 1st camera can capture intruder correctly. No role/information passing occurs when person changes facial direction. Hence 2nd camera can not immediately start focusing after face direction change. Video
Only Cooperation "Idealistic notion". All cameras assumed to be equally capable of undertaking the focusing task. Cameras cooperate by giving up their role to the other after a fixed duration. This is not optimal as the roles are being transferred to non-suitable sensors. Video
Coopetition + PID Roles and information correctly passed when person changes direction. The feedback method being employed is Proportional Integral Differential (PID). This seems good but performance can be further improved. Video
Coopetition + MPC "Our approach". Roles and information correctly passed. Analysis showed average 10-15% improvement compared to PID feedback in terms of number of facial images obtained and their size. Video