Options
1995
Doctoral Thesis
Titel
Ein Ansatz zur Formalisierung der Echtzeit-Bildauswertung und seine Nutzung zur automatischen Erzeugung lauffähiger Maschinenprogramme
Abstract
In applications of real-time image processing, images must be processed pace-keepingly with external processes. This leads to high data rates which require special real-time image processing devices. Such devices consist of configurable pipelines of dedicated electronic processing units, programmable processors, and data buffers. So far, these complex devices must have been programmed manually at a device-near level: The image processing algorithm must be decomposed into processing steps, the processing steps must be scheduled, the requested functions must be assigned to the given physical resources, and numerous operational parameters must be set. This manual "flow control planning" has several drawbacks: It is time-consuming and error-prone, it can be done only be experts who have detailed knowledge on the target device, and the resulting programs and restricted to only one specific target configuration and are thus badly re-usable. In order to overcome the drawbacks of a manual plan ning, an automatic flow control planner has been designed and implemented. Using this flow control planner, image processing algorithms can be specified in a device-independent, functional way by means of data flow graphs. These specifications are then automatically transformed into a control program for a desired target device. A graphical documentation of the planning results is created automatically as well. The programming environment is not restricted to one specific target device, but it can be applied to any real-time image processing device which has a configurable pipeline architecture. In order to verify that an automatic flow control planning can be successfully applied to real-time image processing applications, a case study has been performed. For a set of industrial applications which have been programmed manually in the past, the automatic planning and the manual planning are compared concerning the quality of the results and the time spent. It is shown that the results of the automatic planning are comparable to those of a manual planning by human experts, but they are created in much less time. Therefore, the automatic flow control planning offers a visible improvement of efficiency for the programming of real-time image processing devices.
ThesisNote
Karlsruhe, Univ., Diss., 1995
Language
German