In the first part of this talk the past and current developments in the field of video content analysis (VCA) are discussed. A number of statements are made emphasizing the need for fundamental VCA concepts and principles, based on which the VCA solutions can be developed that are characterized by more robustness, reliability and much wider applicability than the state-of-the-art VCA solutions. In particular, the ``curse of domain knowledge'' and the lack of creative, non-conventional, out-of-the-box ideas addressing the true nature of the VCA problems to be solved, are identified as the main bottlenecks preventing the further growth in the VCA research field. 

In the second part of the talk the research at Delft University of Technology is presented, that aims at a generic framework for affective video content characterization. The framework is based on the valence-arousal paradigm that is used to uncouple the feature measurements from the semantics (affect) inference and provides the possibility to elegantly bridge the semantic gap without the need for explicitly modeling the abstract and ambiguous affect categories as required by the classical approaches based on supervised classification. Application potential of the proposed framework is illustrated in the context of personalized video content delivery and video highlights extraction.