Car detection. Best way. Like FreeFlow

asked 2014-05-19 08:12:34 -0500

gabrielpasv gravatar image

updated 2014-05-19 08:20:39 -0500

berak gravatar image

Hi Guys,

I want to write a software where it will have a IP camera as input and I need to detect cars and take a snapshot of the front of the car and then send this image to my OCR. I already have a OCR library to integrate to my code.

The cars will be passing by a detection area and I should be able to take the snapshot. How do I do that ? template matching with every frame of the video ? How do I add more then 1 template to compare on each frame ?

Sometimes the car will be passing fast and sometimes slow. I need something similar to the freeflow. See this video https://www.youtube.com/watch?v=5Kfp7EFmT98

I already have the OCR which needs a image as input. So now I need to find a way to do the detection.

Thanks

edit retag flag offensive close merge delete

Comments

maybe you don't even need to detect cars, but number plates only

also forget template matching, won't work in your case (not every car is the same, scaling involved, etc)

berak gravatar imageberak ( 2014-05-19 08:16:21 -0500 )edit
Balaji R gravatar imageBalaji R ( 2014-05-20 04:27:57 -0500 )edit

i didn't express myself correctly. I already have a ANPR library (Automatic number plate reader.) I just need a way to capture the front of the car from a video stream so I can send the snapshot to my library to read it. I already have a plate recognition system however it needs a image as input. So now I need to find out how to capture the correct frames.

Thanks for your help.

gabrielpasv gravatar imagegabrielpasv ( 2014-05-21 06:13:53 -0500 )edit
3

all i'm saying is: finding rectangles, filtering them for size and checking good candidates against an svm works nice and fast in the code above.

explicitly searching for car fronts might be overcomplicating it.

berak gravatar imageberak ( 2014-05-21 06:42:40 -0500 )edit