| OCR Text |
Show 43 a "system-specs" FROB class. Each instance repres nts an appli ation rn h d which the selection rules deem appropriate for the requirements. The slot valu s of the system-specs instance represent parameters which determine how the model and application system will be created during training. Figure 5.3 shows the system specifications defined for the requirements in the previous figure. These values are defined by both the application selection rules and the application specific rules which begin the training process. When the necessary parameters for the application system have been defined, the training process begins. Training is performed by a Lisp function which can be done manually, as is for system testing, or by an application specific rule. The trainer uses the parameters defined to perform its task and produce as a result two FROB instances, one a model class instance FROB containing model information and the other a logical sensor instance representing the application system. The system specification instance seen in figure 5.3 was created with parameters to match the high speed, low accuracy values defined in the requirements. The number of application specific rules needed for an application depends on the constraints of the application implementation. For the global feature application there theoretically exists more flexibility in what features can be used as well as how the model can be configured. In practice however the overrobustness of the Area feature made the recognition application of this method inflexible. Local feature focus on the other hand was relatively well constrained with flexibility only in its tolerances for prototyping, clustering and recognizing local features. In practice these parameters proved highly useful in adapting the application method to the different requirements. Experimentation with the recognition system was necessary to gain the knowledge needed for the application specific rules. This is an ongoing process as the system evolves, with the user providing the learning element for the system. The rule base also changes as new applications |
