This paper deals with the supervision of perception tasks, as a component of an autonomous system. First, the role of this component in an autonomous system is presented. Secondly, a model of supervision of perception tasks is proposed : needed control mechanims as well as knowledge modelling are detailed. Then, an implementation of this model, designed as an expert system shell, named OCAPI is presented. The facilities provided by OCAPI are shown, through an example : the supervision of an object detection process in road scenes. More precisely, the expert system, developed with OCAPI, automates planning (selection of a sequence of programs) and control of execution (parameter initialization and adjustment as well as result evaluation). The contents of the knowledge base and particular utilization of this knowledge base are described.