HQ Trivia Helper

Inspiration

“HQ Trivia” is a live mobile game show available for iOS and Android. It's hosted everyday at 1pm and 8pm CST. The players must answer all 12 questions correctly to win. When people first hear about this game their first question is "Can't you just google the answer?" Unfortunate, players only have 10 seconds to answer each question. It's near impossible to type into google by the end of the 10 seconds. That's where my program comes in.

What it does

HQ Trivia Helper reads the question off the screen (phone screen is casted to on my desktop) and googles the result. My current implementation googles each answer with the question attached and returns the number of matching results. (Takes about 2 to 3 seconds!)

How I built it

I used OpenCV to parse images off my phone screen. It parses 4 total images, 1 for the question and 3 for each answer. I then run Tesseract OCR, a text recongnition engine, to convert each image to a string. I then use a combination of the question and answer as a search result using Google Custom Search API. I used Samsung SideSync to cast my phone screen on my desktop.

Challenges I ran into

My model was not reliable when giving the answer due the the nature of the web. Here is an example: Q: What is the very last car on a train often called? A1: New York Jets Car (Highest Results) A2: Train drain A3: Caboose (Correct Answer)

My implementation returned A1 and A2 having over TEN times as many results as A3. I believe there are lots of article over New York on the web which causes an inflation of A1 results. A2 just adds the word “drain” (“Train” is a repeat) so it’s still a vague search.