Wheel of Fortune Call Letters Computer Vision Project

I am an avid fan of Wheel of Fortune. I have often wished that there was a full list of the letters that have been called. It is easy to lose track of what has been called already that was not part of the puzzle. This dataset is part of a project that provides a simulated puzzle board and call letter list on a web page that gets updated in real-time as the letters on the show are called.

The dataset contains 4,597 images of the call letters that appear on a Wheel of Fortune screen when players guess a letter. The majority of these have been generated. I managed to get a few hundred screenshots from the show, but that took considerable time and didn't give me nearly enough letters to traing with - especially the rare letters. By generating my own images, I was able to ensure all letters were well represented and since I was placing the letters myself, I knew the location and size and could create the annotations at the same time.

You will observer that the images are only a portion of the screen from the show. When I am detecting letters within the images, it was easier to isolate the call letters. This approach gave me far more accurate results and my project code crops that section of the screen to send for inference.

I decided to split out the puzzle board letters into a different dataset. After testing, I found this was a better way to get a high level of accuracy in letter detection. You can find that dataset at the link below.

https://universe.roboflow.com/warren-wiens-d0d4p/wheel-of-fortune