Rapid prototyping in the development of image processing systems

Abstract

This contribution presents a rapid prototyping approach for the real-time demonstration of image processing algorithms. As an example EADS/LFK has developed a basic IR target tracking system implementing this approach. Traditionally in research and industry time-independent simulation of image processing algorithms on a host computer is processed. This method is good for demonstrating the algorithms' capabilities. Rarely done is a time-dependent simulation or even a real-time demonstration on a target platform to prove the real-time capabilities. In 1D signal processing applications time-dependent simulation and real-time demonstration has already been used for quite a while. For time-dependent simulation Simulink from The MathWorks has established as an industry standard. Combined with The MathWorks' Real-Time Workshop the simulation model can be transferred to a real-time target processor. The executable is generated automatically by the Real-Time Workshop directly out of the simulation model. In 2D signal processing applications like image processing The Mathworks' Matlab is commonly used for time-independent simulation. To achieve time-dependent simulation and real-time demonstration capabilities the algorithms can be transferred to Simulink, which in fact runs on top of Matlab. Additionally to increase the performance Simulink models or parts of them can be transferred to Xilinx FPGAs using Xilinx' System Generator. With a single model and the automatic workflow both, a time-dependant simulation and the real-time demonstration, are covered leading to an easy and flexible rapid prototyping approach. EADS/LFK is going to use this approach for a wider spectrum of IR image processing applications like automatic target recognition or image based navigation or imaging laser radar target recognition.