Loosely-Coupled Telepresence Through the Panoramic Image Server

While computer vision systems can clearly assist in surveillance tasks, taking the human out of the loop entirely proves to be di cult or undesirable in many applications. Human operators are needed to detect events missed by automatic methods, to identify false alarms, and to interpret and react appropriately to complex situations. A key challenge in partially-automated systems is how best to combine machine algorithms for event detection, analysis and tracking with telepresence systems controlled by one or more human operators. Given the disparity in performance between the human visual system and typical robotic cameras, we argue that direct coupling of human and machine is inappropriate. We propose instead to couple human and machine components indirectly, through a database called the Panoramic Image Server. We show how this loose coupling allows machine and operator surveillance priorities to be resolved while providing a fast and natural telepresence environment for one or more human operators.