In our limited domain of machined features, machining constraints can be
exploited to identify curves as machined features. For instance,
if a negative feature contains a positive feature, then it must be
an island within a profilePocket. If necessary, the
island
may be trimmed to the sensed height with a negative feature such as the
profileGroove. If a negative feature contains no positive feature
and is composed of a arc segment, then it may be represented
by a hole. If a hole contains no other
features, and the interior of the raw image is below a threshold,
it may be considered to be thru.
Some aspects of machined features are difficult to measure accurately using our current image processing algorithms. For example, fillets between line segments in a profilePocket may not be within the accuracy of our vision, but are necessary for machinability. In cases such as these, default values are selected so as to have minimal deviation from the reverse engineered model, yet allow the model to be machined. It is anticipated that by recognizing likely locations for such difficult aspects, they may be explored effectively using a CMM (coordinate measuring machine) or another probing device.