Active video analytics one leap ahead applicability and entering new dimensions

摘要

How to overcome the gap between people's expectation on capabilities of video surveillance systems and perceived performance? One solution is to focus on specific, well defined and tailored problems as well as well-defined operational conditions. But how to restrict the reality? While some specific problems are generally addressed and broadly covered within the community and thereby partially solved at least, general performance for unrestricted environments is still difficult to obtain. Thus we see the need to seek for general applicability in order to enable new application domains. We propose a paradigm shift from fully pre-configured solutions towards solutions that learn and adapt on-the-job autonomously and/or being interactive by responding to operator feedback in real-time. In that way such systems are error-tolerant and responsive and ease end users expectation by improving in performance over time. Following that mindset we will present some selected research topics we are currently working on. In addition we will briefly present undergoing work to leverage two relatively novel paradigms in the field of video analytics, namely data acquisition by a moving camera (e.g. Micro Arial Vehicles -MAV) and 3D+t data reconstruction in conjunction with change detection.