Object Recognition is to automatically detect certain objects by a computer.
People recognize objects without much effort, although objects rarely appear exactly the same. Sometimes we stand close to or far, which causes the object to look bigger or smaller, in other places the same object is rotated or hid after something else. Nevertheless, we have no problems recognizing this, computer-controlled object recognition makes this a big problem. A pioneer in object recognition was David Lowe, a Canadian scientist. He came up with the idea of taking an item that does not change through translation, rotation, size and slight distortion. The algorithm used for this is called Scale-invariant feature transform (SIFT). Although there are many ways to do object recognition, this is the most common method. The SIFT method
This method of object recognition is done in a number of stages. If we want we to recognize objects, we first need to ensure that there is a database that already contains a number of objects that can be used as a . This database is filled with SIFT key features of the objects that we will use as a . The location and orientation of these features are kept separately so that when comparing to a new object, not just the feature itself must be correct, but the orientation and location must be correct.
When an object is recognized, important features are also taken with the SIFT algorithm, which are then individually compared to all features in the database. The algorithm used to determine whether the attributes match is the Best-Bin-First algorithm. Of all the corresponding features that come from this, only those who maintained the same location, orientation and scale will be .
The results we have now gained are clustered, that is, we now look at the different characteristics together and no longer individually. The matches that previously occurred are grouped by object, where we meet three or more matches per object, others are omitted. These are then more detailed and omitted if they do not match. The final result is a match that we can almost certainly say is the object we have for us. Other methods
wiki
