How to improve Vehicle Color Detection accuracy?

asked 2016-07-03 23:11:45 -0500

damiya14 gravatar image

updated 2016-07-04 00:21:55 -0500

I am doing a vehicle color detection. i have implemented a simple method. Steps are below.

  1. Getting the ROI ( above the license plate on the vehicle bonnet area) (Extracted Image Window)
  2. Getting the HS histogram of ROI
  3. Histogram Back-projection ( BackProj Window)
  4. From the Back-projection area extracting the vehicle image (devide Window)
  5. Calculating distance of HS values from predefined HS values for colors. (Doing this for each and every pixcel on the vehicle area shown in devide Window)
  6. Voting the color and getting the color which has the maximum number of votes.

From thees steps i have only achieved an accuracy of 75%. How can i increase the accuracy? Is there any other method to do this?

I am very new to opencv :) :)

image description

edit retag flag offensive close merge delete


You did not normalize somewhere for the current lighting conditions, so comparing colors will always be biased ... not sure if you can get a higher accuracy if you don't do this.

StevenPuttemans gravatar imageStevenPuttemans ( 2016-07-05 04:30:04 -0500 )edit

please explain how to normalize as for the current lighting conditions. i am very new to this.

damiya14 gravatar imagedamiya14 ( 2016-07-05 22:35:00 -0500 )edit