Hi, for my senior design we have are using the YOLOv3 network for objection detection of the game of casino game of Roulette (specifically identifying the 0 pocket and Roulette ball). We already have annotated those images (~2,000ish) data set at this point and our getting decent object detection results. We also have coded functions that compute speed (pixels/second), radius/distance (pixels), angular velocity (degrees/second), theta (degrees), acceleration (degrees/second) per frame and output that data to a txt file ( can be CSV if needed). I was figuring that I could use a predictor neural net for like predicting Weather, Stock Market or House prices, since the general concepts seem the same.
Here’s a sample : https://www.youtube.com/watch?v=HaMlKvNqCVs
Beating Roulette
I thought outputting the parameters of each frame, like the initial speed/velocity of the ball, and where it finally lands (in what pocket) in a python data frame and run some Regression on it could produce “decent” results, i.e.
speed_initial velocity_initial acceleration_initial theta_initial pocket_final
500,454 400 433 308 17
..... ......... ...... ...... .....
Obviously I would need a lot of data, but I thought the neural network would be able to build a model of what entry speeds and angles result in landing in some specific pocket.
Are there any better approaches? My teammate was planning on using a binary classifier to divide the wheel in 2-halves and the output parameters (speed, accel, theta, radius , etc…) into a binary classifier as a tensor…
If anyone has any suggestions please me know! I know there’s experts on these forums that may have done similar prediction models that maybe i can tweak/twist to fit this roulette prediction!