| OCR Text |
Show 56 y Orient Vector Chord Chord 2 Corner Location: (x,y) Included angle: 90 X Figure 5.9: Corner Detection in an Object values assigned to them work best. If every change in chord angle from negative to positive were reported as a corner, many false corners result from a camera image due to noise and quantization errors. A tolerance was set for how much the angle between the chords must change before a corner is to be detected. Another tolerance was set for the chord angle itself; this eliminated detection of smooth curves, only angles less than the tolerance are reported as comers. These tolerances were adjusted to remove most of the false comers detected in the camera images. The rate of false detection with the tolerances used was on the order of 1 in 15 which is about 7 percent, an acceptable rate. Figure 5.10 shows an image of a CAD designed object from which holes and comers have been extracted and labeled for use by the model builder. 5.2.4 Building a Local Feature Focus Model The model building part of the local feature focus application was written in Lisp. Lisp functions perform all the training for the application. This involves the |
