Ask Your Question
1

Bottle recognition on predefinited enviroment

asked 2014-05-26 15:21:06 -0600

ciclonite gravatar image

Hi all, i'm a newbie with opencv and i need help for my project. The main goal is recognize some types of bottles in a controlled enviroment. The controlled enviroment is a white or black box where a bottle will be put inside. The light are costant over the time. If i put a generic bottle inside ( 0.5lt or 1,5lt or 2lt or other type) the algorithm should recognize the bottle, if i put other type of object should return an error. What kind of algorith can i use as starting point? The HW that i use is a simple beaglebone black, thus i need an algorithm that is suitable for this hardware. Thanks a lot. Giovanni

edit retag flag offensive close merge delete

Comments

Post a couple examples of bottles you want to recognize. Do you want your algorithm to have binary output: bottle - no bottle or do you need to recognize the shape of the bottles precisely (so as to reject similar but not identical bottles for example)? What if someone puts there a bottle of ketchup?

Witek gravatar imageWitek ( 2014-05-26 19:02:14 -0600 )edit

thanks for reply witek, I need to recognize the color, and if the bottle is present or not. For the type of bottle I thought of a database containing the keypoint for the bottles to compare.

ciclonite gravatar imageciclonite ( 2014-05-27 04:49:27 -0600 )edit

Is it possible that someone will insert a can or a jar in your machine and you need to recognize this situation?

Witek gravatar imageWitek ( 2014-05-27 12:32:08 -0600 )edit

yes it is possible Witek. I need to recognize this situation.

ciclonite gravatar imageciclonite ( 2014-05-30 14:18:07 -0600 )edit

2 answers

Sort by ยป oldest newest most voted
1

answered 2014-05-27 05:29:39 -0600

FLY gravatar image

updated 2014-05-27 05:30:21 -0600

First of all if the environment is constant you better to use SVM approach with some feature detection algorithms or object detection , you can use many approaches for it , like SURF , BOW , BOF , SIFT etc , and for other approaches you discuss like color detection in them etc you can simply can get the idea of it from these sources

Source1

Source2

Source3

edit flag offensive delete link more
0

answered 2014-05-31 15:56:22 -0600

Witek gravatar image

Since many bottles are quite similar and/or transparent, I think that using keypoints is not a good idea. You will get many edge keypoints that would look exactly the same on each bottle. However, keypoints would be perfect if you were to identify bottles by labels on them, but since you did not mention it, I think this is not the case. Therefore, my suggestion is to subtract the image with a bottle from the empty background. That (provided you have proper lighting generating no shadows) would give you a nice silhouette of each inserted item. Find the contour of the item and compare it with the contours stored in your database. Use the width and height information, matchShapes function and perhaps number of vertices of the contour approximation.

edit flag offensive delete link more

Question Tools

Stats

Asked: 2014-05-26 15:21:06 -0600

Seen: 1,813 times

Last updated: May 31 '14