To
teach a computer to learn an object, we could let it taking images of sample objects
and stored in database.
For
example, taking photo of a hot dog with maximum and minimum sizes recognized
for a hot dog. The max and min size would properly be scaled in sample images
stored in database. Of course, operators must enter the real max/min size of
that hot dog in database.
Assuming
we take photo of an object (hot dog) at the same distance as when taking
sample, the photo of an object (hot dog) with appropriate photo scale could be
recognized as a hot dog or not. With max or min sizes’ photo in database, we
could estimate the size of the captured hot hog. If an object is larger than a
max hot dog, system would rule it out.
Speaking
of pattern recognition, we could think the sample object as a transparent. When
an object is captured in photo, we could apply the transparent of sample
objects on top of the captured image. Operators must train the system to
determine some acceptable variation of the captured image. If variation conditions
of the sample image and captured image are met, that object is identified.
We
could think of an object is composed of curves, straight lines, etc. We could
zoom in or out of sample image on top of a captured object to compare its shapes
in terms of curves, straight lines, etc. Some curves of the captured image
could be varied a little bit as compared to the sample object. (Operators must
train the system to determine some acceptable variation of curves and lines.)
If
we take photo of an object at the same distance, we could estimate size of
another object placed at the same distance.
If
we know the size of an object, we could estimate the distance between the
camera to the same object placed at different distance.
If an object image is captured at a known zoom, the real size of that object could be estimated as well as the distance between the object and the cameras, i.e. the size of a photo image must be a constant in width and length. For example, at a certain zoom, system would be able to tell the real size of the scene in a photo; so an object’s real size and its distance could be estimated.
2.
Voice pattern recognitionIf an object image is captured at a known zoom, the real size of that object could be estimated as well as the distance between the object and the cameras, i.e. the size of a photo image must be a constant in width and length. For example, at a certain zoom, system would be able to tell the real size of the scene in a photo; so an object’s real size and its distance could be estimated.
When we talked, our voice was captured as waves in an image. If we used the same strategy as in image recognition, we could determine correct words, letters, or phrases.
No comments:
Post a Comment