In the context of video indexing, we present the Clavis system in which typical video sequences of television programs are represented by templates. Templates are terminological constraint networks in which video segments coming from automatic analysis tools are represented in a description logic formalism. Templates allow to express complex classes of video sequences with temporal constraints associated to a regular expression operator. Recognizing occurrences of a template in a video program is a plan recognition problem for which efficient methods have been implemented in a constraint satisfaction problem framework. The paper describes the system and illustrates its use with several experiments that were done in the context of the DiVAN european project.