how to train input in nn
Asked by Tulips on 14 May 2012
Latest activity Commented on by Tulips on 15 May 2012
hi Mr expert, can u explain me how to train inputs in neural network?im currently working on road signs recognition. Do I need to compile a database of road signs ? i simply cant understand the ways. Im not ask u the method to recgonize road signs but asking u the ways.pls explain in details.Thank you for spending ur time explain it.
Products
No products are associated with this question.
1 Answer
Answer by Walter Roberson on 14 May 2012
Accepted answer
You do not need a database of road signs. You can use instead a description of exactly what to look for in each sign to distinguish it from all the other kinds of sign.
For example, "eccentricity 0 (circle), outer circle 2% of sign diameter, outer circle is green, inside outer circle is unbroken white" might perhaps be the description for a "parking is allowed here" sign. The description for a sign that indicates "no left turn on weekday mornings" might be much more complex, especially as it might occur in the negative (left turn crossed out) or in the positive by omission (only straight ahead is shown on sign.)
If you have an exact description of each road sign, you might still need a NN to figure out whether the street is washed on the 16th of even numbered months or of odd numbered months.
Individual aspects such as a colored circle around the outside of a sign are known as "features", and it is usually easier to describe each sign by its features and then use some kind of processing routine (such as a NN) to figure out what to look for to distinguish a particular sign from all other similar signs.
5 Comments
Show 2 older comments
Tulips on 14 May 2012
ya, thx sir, then cud I simply train the pixels of the bounding box in NN toolbox?
Walter Roberson on 14 May 2012
Not unless you don't mind confusing camel crossing signs with elk crossing signs.
http://en.wikipedia.org/wiki/Traffic_signs
0 Comments