Having made it to the 2017-2018 FIRST Tech Challenge world championship, we needed to up our game for match data collection. At previous competitions, we wrote notes on paper worksheets and then tabulated the data by hand in a spreadsheet. This was a time consuming system, and such significant manual data entry was just asking for major mistakes and errors. We needed a new system. Over the days leading up to the competition, I built a computer vision system for tabulating data from scantron-like bubble sheets entirely from the ground up using only minimal camera and image processing libraries limited to just reading and writing the values of pixels. I implemented a custom filtering and target identification system capable of accurately and reliably reading bubble sheets, even in uneven lighting and with very low image quality. At the competition, it successfully tabulated more than 5,000 data points in total from hundreds of data sheets. This significantly reduced our turnaround time for match data while increasing data input accuracy.
You can see some highlights of the system below.