Ask Your Question

[SOLVED]Determining number of Traffic Light Pattern after Detecting Traffic Light [closed]

asked 2019-10-04 01:04:04 -0500

trohit gravatar image

updated 2020-05-22 08:27:02 -0500

supra56 gravatar image


I am working on a project for detecting traffic lights in real world environment. I have trained a Haar Cascade classifier which tries to find Traffic Lights in a pre-defined Region Of Interest(30% from left and right is cropped and 10% from top and 50% from bottom). I use the color segmentation to find which color is present on detected traffic light and so far it works fine. However, i have a problem in finding the patterns of traffic lights, i.e. there are cases of left or right arrow present rather than just red and green colors. Please check the attached images for better understanding. We also used the Aspect Ratio of a detected traffic light (by detecting contours from a binary image) to differentiate between 3 pattern and 4 pattern of traffic lights but it doesn't work as expected. Could you suggest an efficient way of distinguishing 3-bulb traffic light from a 4-bulb one? I am trying to port it on an embedded system, so my computational resources are very limited. Thank you in advance. C:\fakepath\1.png


edit retag flag offensive reopen merge delete

Closed for the following reason the question is answered, right answer was accepted by supra56
close date 2020-05-22 08:26:25.409779


Finding a contour seems excessive, especially that from my understanding the detected will be very small on the image, thus it will be hard to distinguish 3 from 4 patterns. Can you share the image of a contour detection?

richardo gravatar imagerichardo ( 2019-10-11 03:48:00 -0500 )edit

1 answer

Sort by ยป oldest newest most voted

answered 2019-10-11 03:43:03 -0500

richardo gravatar image

Have you tried to use any methods for detecting the arrow? If you know that the arrow is present, that means you detected a 4-pattern traffic light. Otherwise, it doesn't matter if it's 3 or 4-pattern, you will just use color segmentation as you did before. Hope it helps.

edit flag offensive delete link more

Question Tools



Asked: 2019-10-04 01:04:04 -0500

Seen: 289 times

Last updated: May 22 '20