Advances in Computer Vision are making it possible to move forward from the design of algorithms used punctually or in visual servoing, to the conception of complex applications containing aspects such as scene analysis, decision making, supervision and real-time processing. This increased complexity reflects upon all the levels of the development process of these applications, namely the design, validation and implementation. This work proposes a design framework for the implementation of real time vision applications, in addition to a variety of adequate tools. The proposed framework is based on a functional structuring of the required processing into elementary tasks, followed by their logical composition in a order to build the vision application. These elementary entities are called Vision Tasks. They are seen as logically controlled, parametrized real time processing loops capable of controlling a visual sensor. The Orccad/MaestRo environment, developed initially for robotics control, exhibits interesting characteristics for its use in the proposed framework. Specifically, it enables the use of tools for logical control validation and allows efficient management of the real time resources. However, and in order to draw full benefit from using this environment for developing vision applications, we expand its functionality. Major modifications encompass an adaptive dynamic parametrization mechanism, a communication scheme that ensures the interoperability of the system and automatic generation of supervision interfaces. An application of the proposed methodology and of the use of the modified Orccad environment are presented for a typical tracking application in a context of assistance to elderly persons.
