Second, defining areas or objects of interest by
solely relying on computer vision might be hard to
achieve in the near future. Arbitrary user-defined
queries (for example, searching all cars in the
videos of the database that received the user’s attention)
are required to process the recorded data
to its full extent. Semiautomatic approaches and
crowdsourcing could bridge the semantic gap in
automatic approaches. Hence, visual analytics
could help support such semiautomatic analysis.