Real time image detection (Python)

asked 2016-08-31 05:50:23 -0500

htn gravatar image

Hey guys, I wanted to ask which methods would be ideal for finding a certain image via a camera. I started OpenCV 2 weeks ago so am still relatively new to this but have looked into using Multi-scale Template matching with little success. The code that I have been fiddling with comes from this website link text

Any help would be greatly appreciated, thanks!

edit retag flag offensive close merge delete



and your question is ?

berak gravatar imageberak ( 2016-08-31 05:57:46 -0500 )edit

my question is what would be the best way to detect a certain image (so like a sign) via a camera and to draw a rectangle around it once it has found it

htn gravatar imagehtn ( 2016-08-31 06:03:17 -0500 )edit

can you be more precise, on what you are really looking for ? also, an example image might be helpful !

berak gravatar imageberak ( 2016-08-31 06:11:21 -0500 )edit

I guess what I'm trying to do is template matching an image via a camera. So I will scan an image of a game's cover and the output will be the name of the game. image of game

htn gravatar imagehtn ( 2016-08-31 06:17:35 -0500 )edit

well the thing is, it highly depend on the type of object. For example a traffic sign, will require a different approach then your game cover. So be as specific as possible to the actual thing you want to find!

StevenPuttemans gravatar imageStevenPuttemans ( 2016-09-01 04:59:20 -0500 )edit

@StevenPuttemans - I'll try to be as specific as possible - I'm working on a Robot using OpenCV that reacts to signs (Go, stop, Turn left, Turn Right, 360 turn, etc). So when the robot "reads" one of these signs it will know: a) what the sign is and b) what method to execute. My problem is I'm not entirely sure how to get the Robot to recognise signs via a camera - I've tried multi-scale template with no success (I was told that multi-scale template matching would be best for a project like this).

htn gravatar imagehtn ( 2016-09-01 06:07:54 -0500 )edit

Can you add a print of your signs? Because if they are fairly simple, you could train simple cascades with few levels, which run very fast over lower resolutions, which is mostly the case in robotics! Or how about using aruco markers and select a specific marker to identify a movement?

StevenPuttemans gravatar imageStevenPuttemans ( 2016-09-01 06:27:54 -0500 )edit

Hey @StevenPuttemans - Thanks for the speedy reply! I've just had a little look into Aruco markers. How robust would you say this is against false detections? With my current signs I've tried an approach with just detecting squares but got a lot of false detections! I think this marker detection thing might be more reliable?

htn gravatar imagehtn ( 2016-09-01 06:41:30 -0500 )edit

yep, we have several setups where false positives are non existing!

StevenPuttemans gravatar imageStevenPuttemans ( 2016-09-01 07:24:39 -0500 )edit

Hey StevenPuttemans - firstly sorry for this late reply. Will I need to convert my signs into marker form? Or are these markers 'set' already? Also do you know here I can find a good tutorial so I can have a good stab at this? Thanks in advance

htn gravatar imagehtn ( 2016-09-05 05:11:45 -0500 )edit