Created using C++ with OpenCV.
Program was first trained of a series of negative images. After training the program can detect dartboards in images. See the included report for a full writeup.
The Thresholded Gradient Magnitude, a 2D representation of the Hough space and the result of the final detection for two example Dartboard images.
The following Flow diagram shows how we combined evidence from the Hough Transform and Viola-Jones detector.
- We used the Viola-Jones detector to generate boxes around dartboards and other similar images. This produced lots of boxes and a high False Positive Rate (FPR).
- We combined this data with the Hough Circle Transform to remove boxes that didn’t contain a centre of a circle, because we found dartboards were usually round.
- However, to catch slanted dartboards (dartboards that were not a perfect circle) and dartboards that were partly visible we needed to combine evidence from the Hough Line Transform because we found line endings usually clustered around the centre of a dartboard.
- We kept a box if it contained a centre circle, or if it contained more than 5 clustered line endings, to get a lower FPR.