ChessCam | Convert your OTB games to PGN! WOW! 🤳♟️➡️📃😲 Chess Chats #9
#chesscam #computervision #otb2pgn #chesschat
Have you ever wished that you could easily get a PGN of a game of chess you played over-the-board?
I know that I have! Yes, I could have simply written it down during the game, but that's kind of a hassle when playing a casual game with a friend at a cafe or bar. Recently, I've been using an electronic chess board like the Chessnut Air or Chessnut GO. However, sometimes I like to bring some large tournament sized pieces, and a fold up silicone board as there's something fun with just larger pieces!
Earlier in the year, I discovered a project run by some Australian engineers to create a standalone device, the Clio ɑ that uses computer vision to create a PGN of an OTB game. Unfortunately, the team members have real day jobs, and it seems that the project is advancing slowly.
So, I was very excited to discover last week an absolutely amazing app/web service called ChessCam (https://www.chesscam.net/), created by a member on Lichess by the name of BlindfoldBlunderer. It's available completely for free online and also as an Android app (though this seems to be simply a wrapper for the website itself). The service (which I thinks runs locally on your device/computer, so no data is sent to their servers) uses computer vision on a video (or a live feed from your device camera) to "read" the board, and automate the recording of the moves to a PGN directly!
Today, I gave the app a go using three different video recordings of the same game, using (i) a mobile phone (Samsung Galaxy S24 Ultra) raised to give an overhead view, (ii) an action camera (DJI Osmo Action 3) giving a near, but wide-angle recording, and (iii) a more distant wide room shot on my main video camera (Fujifilm X-S20 + Viltrox 13 mm). I recorded two short games, the von Hennig Gambit "oh no, my Queen!" Trap with Bxf7#, and the Englund Complex Trap Line with Qc1#.
The results?
Curiously, ChessCam made some mistakes with the Samsung Galaxy S24 Ultra recording, even though (to my eyes), this had the "cleanest" view of the board. On the von Hennig Gambit game, it didn't capture Black's final move (Bxd1) or the checkmate. Worse, on the Englund Gambit game, it incorrectly recorded the position of White's queen move.
ChessCam's computer vision system couldn't identify the (smaller and more distant) board on the Fujifilm X-S20 footage.
At this point, I was a little disappointed, but was then rather surprised by ChessCam's performance on the DJI Osmo Action 3's footage, which gives a rather distorted view given the near distance and wide angle. It managed to create a PGN of the von Hennig Gambit game perfectly. Not shown on the video, it did have trouble with the Englund Gambit game.
My conclusion?
You can't beat free! And I'm thankful that there are talented people releasing chess tools for free to the community! Clearly ChessCam does needs some additional work to make it more reliable. However, I'm going to test it further, possibly with some slightly smaller chess pieces on the same board, as I suspect that some of the difficulty with the detection might have had to do with pieces obscuring each other.
If this is something that tickles your fancy, give it a go and write a comment on your experience!
