Predictor bot / video capture & machine learning

[Robotositos]

Hello everyone,
Here is a post to show you my current predictor bot using video detection.

The concept of my system is to record and capture data from spin like Zero/Ball/Pivot  XY position, speed and store it into a database. After that, the next spin will be compare to the old spins stored in the database and ouput a "predicted result", The predictor will search for the "nearest sequence" in the database. I managed to get a database with 20k spins in a week before I notice the casino move the "pivot" wich invalidate the old data.

Here is an example of data for one spin :
_{[[20201008171629], [436.13, 257.92], [98.68], [50], [478.81, 222.61, 478.81, 222.61, 478.8, 222.59, 478.82, 222.79, 478.79, 222.82, 478.79, 222.82, 478.47, 222.72, 477.11, 222.73, 475.06, 222.23, 471.44, 221.45, 467.14, 221.05, 462.51, 220.77, 457.36, 219.79, 242.32, 139.57, 443.2, 218.85, 434.51, 218.31, 424.91, 218.78, 414.06, 220.01, 403.2, 220.13, 391.44, 221.9, 378.71, 223.59, 366.13, 226.66, 352.87, 229.8, 352.87, 229.8, 338.51, 233.79, 323.68, 239.04, 295.09, 252.75, 295.09, 252.75, 280.94, 259.78, 268.37, 267.8, 254.76, 276.42, 242.49, 286.29, 232.43, 297.84, 220.72, 307.76, 210.99, 319.49, 203.0, 332.06, 194.03, 344.59, 187.55, 358.27, 181.82, 372.0, 177.08, 386.87, 173.17, 401.41, 170.4, 415.46, 169.88, 430.71, 169.88, 430.71, 169.57, 446.75, 169.9, 462.64, 171.53, 478.29, 174.25, 494.15, 179.25, 509.65, 184.97, 525.39, 193.21, 538.47, 201.24, 553.62, 209.88, 566.31, 220.41, 579.27, 233.23, 590.73, 245.73, 603.0, 259.67, 613.73, 274.46, 623.1, 289.72, 632.28, 306.28, 639.52, 323.36, 647.38, 340.58, 654.26, 358.04, 658.91, 375.47, 661.86, 394.59, 663.62, 414.75, 665.44, 436.34, 666.33, 455.97, 666.1, 455.97, 666.1, 475.54, 664.21, 497.42, 661.61, 513.63, 657.42, 531.43, 653.03, 549.31, 646.27, 565.23, 638.85, 581.23, 631.36, 596.86, 621.61, 610.7, 611.39, 623.09, 601.2, 636.28, 589.78, 647.18, 577.51, 658.3, 565.25, 667.45, 551.73, 675.09, 537.43, 682.78, 524.6, 687.0, 509.71, 692.34, 494.57, 696.23, 479.15, 698.09, 463.56, 699.13, 448.91, 699.07, 433.04, 698.17, 417.71, 695.94, 403.45, 695.94, 403.45, 692.35, 389.76, 686.95, 375.93, 682.23, 362.27, 674.7, 349.28, 667.37, 335.13, 658.59, 323.12, 649.56, 312.63, 638.9, 300.71, 627.25, 290.1, 616.88, 281.15, 604.59, 270.72, 590.59, 262.95, 578.15, 255.05, 563.78, 248.09, 549.23, 242.3, 534.65, 237.53, 519.5, 232.04, 504.18, 227.59, 488.79, 224.61, 473.27, 221.86, 456.59, 219.64, 438.15, 219.05, 420.24, 219.37, 403.82, 220.91, 403.82, 220.91, 388.13, 222.72, 370.87, 225.29, 355.81, 229.55, 340.52, 233.48, 324.91, 238.97, 309.78, 244.98, 295.99, 252.04, 282.21, 258.84, 269.27, 267.09], [16]]}

[Timestamp], [PivotX,PivotY], [Ball Speed], [Zero Speed], [Zero X, Zero Y....] [Result]

The detector :

some GUI :


All is coded in Python with a realtime object detector call "YOLO"

I managed to get some wins but i long term my predictor fail, and because the casino move the "pivot" each week I can't get more than 15-20k spins in my database. I decide to capture pivot position to keep my data clean but we can get more than 15-20k spins.

What do you think ?
I know its a kind of "brute force" system but I would like to have some of your feedback / question
I should post a video to show you the complete process.

Thanks for your time and sorry for my terrible english.

[Ion2307]

Hello
i am interested in your bot Predictor bot / video capture & machine learning.

[Herby]

Hi Robotositos,

I built a picture of your [Zero X, Zero Y....] positions.

move the "pivot": do you mean what you call "center" in your picture is moved ?
With some math tranformations you could transform the Zero positions to all center positions ?

Do you have the ball positions as well ?

Cheers

[Clf7]

Could you explain what they move? I dont get what the pivot is...Thank you

[Roulettebeater]

Looks good - I would help you with data Analysis  only if you can explain more what u mean with casino changes the pivot.

[Robotositos]

Hey everyone,
What I call Pivot or Center is the huge metalic thing at the middle of the wheel.
Each week the pivot move a little bit, I assume they move the inclinaison by some degree to add more randomness and unpredictable.

Here an example between 2 different week:



You can see the "diamonds" are perfectly in the same position, pockets and numbers are also seems to be in the same place. The camera angle is not changed.

I don't know if this thing can cause bad predict but in case of I prefer to keep my data clean and split it by "pivot" position.

Hi Robotositos,

I built a picture of your [Zero X, Zero Y....] positions.

move the "pivot": do you mean what you call "center" in your picture is moved ?
With some math tranformations you could transform the Zero positions to all center positions ?

Do you have the ball positions as well ?

Cheers

Hey thanks for the graphics, what tool did you use for this ?

I'm not capturing the ball anymore because of multiple factors :
- The ball disappear half screen (capture is slightly the same each rotation)
- The video sending by the casino is capped at a fixed relatively low frame rate causing some "duplicate position registering" and inconsistent capture
- Ball will always start in the same spot in this auto roulette, I only detect the CW/CCW change

Looks good - I would help you with data Analysis  only if you can explain more what u mean with casino changes the pivot.

What I think I really need is a professional of mathematics and physics who can "calculate" a properly prediction with my input.
At this time my project just look at past spin and try to find the "same" in the database... and we all know each spin is independant :p

Thanks for reading !

[Clf7]

At this time my project just look at past spin and try to find the "same" in the database... and we all know each spin is independant :p

Exactly, past spins wont help you

[Herby]

What I think I really need is a professional of mathematics and physics who can "calculate" a properly prediction with my input.

You need the positions of the sphere and the (don't know the right name) vessel (you used the positions of the green zero).
The main forces: gravitation and friction, ... lead to acceleration, deceleration -> different velocities -> changes of positions.
It doesn't matter if you can't track the ball positions for parts of the pictures.

Once ( few years ago) I asked the member: General for the differential equations, he posted kindly a pdf ( I already knew) somewhere in this board.

If you post postions of the ball and the green zero we can do some math work, as far as my time allows.

[Robotositos]

You need the positions of the sphere and the (don't know the right name) vessel (you used the positions of the green zero).
The main forces: gravitation and friction, ... lead to acceleration, deceleration -> different velocities -> changes of positions.
It doesn't matter if you can't track the ball positions for parts of the pictures.

Once ( few years ago) I asked the member: General for the differential equations, he posted kindly a pdf ( I already knew) somewhere in this board.

If you post postions of the ball and the green zero we can do some math work, as far as my time allows.

Here is some capture with my old detector :

Code: [Select]

[834.11, 396.78, 814.41, 329.43, 779.55, 269.23, 733.06, 216.14, 676.8, 170.67, 612.95, 138.08, 540.93, 115.72, 467.1, 103.52, 391.62, 104.17, 317.16, 116.66, 244.01, 141.85, 178.09, 178.63, 120.99, 226.36, 74.63, 284.65, 42.74, 350.74, 27.62, 424.43, 33.71, 499.83, 832.38, 485.24, 835.9, 421.21, 824.8, 359.21, 801.08, 301.84, 765.84, 250.49, 721.19, 204.67, 668.61, 166.19],[220.34, 311.78, 231.86, 300.91, 242.63, 289.8, 254.72, 280.28, 267.69, 271.25, 281.74, 262.42, 295.46, 255.29, 309.98, 248.51, 324.63, 242.62, 340.85, 236.89, 356.53, 232.34, 372.36, 228.86, 388.53, 226.06, 404.54, 223.74, 421.85, 223.33, 440.68, 222.93, 458.43, 223.38, 645.3, 310.1, 655.1, 322.53, 665.16, 334.36, 673.33, 348.03, 681.67, 361.62, 686.51, 376.02, 691.85, 390.81]

Code: [Select]

[834.89, 406.56, 819.36, 341.73, 790.44, 282.64, 747.91, 230.2, 696.3, 185.85, 635.97, 149.11, 571.57, 123.46, 501.38, 107.34, 429.2, 102.89, 356.48, 108.59, 285.46, 126.19, 218.71, 153.72, 158.56, 192.69, 105.43, 242.18, 65.8, 299.17, 38.36, 363.25, 27.46, 434.05, 34.97, 506.32, 824.49, 515.8, 824.49, 515.8, 835.33, 454.61, 834.35, 395.37, 816.75, 336.73, 790.72, 283.26], [643.83, 585.13, 632.05, 597.45, 618.3, 609.02, 604.18, 619.45, 589.59, 628.65, 574.19, 638.01, 557.36, 645.56, 540.67, 652.19, 523.36, 658.63, 506.04, 663.17, 487.4, 666.14, 466.24, 668.43, 447.09, 668.9, 426.56, 669.89, 406.53, 668.47, 387.2, 666.67, 368.89, 664.23, 350.47, 659.64, 184.18, 525.96, 184.18, 525.96, 179.1, 510.85, 174.22, 495.47, 171.18, 481.27, 169.71, 466.27]

Code: [Select]

[834.4, 399.32, 816.77, 336.69, 787.8, 279.73, 746.93, 229.53, 697.13, 186.07, 640.87, 151.82, 577.77, 125.33, 510.81, 108.72, 440.98, 102.65, 372.07, 105.94, 302.53, 120.84, 237.83, 144.7, 176.96, 179.51, 124.29, 222.9, 80.59, 274.94, 48.99, 334.11, 31.06, 397.45, 29.56, 466.79, 44.27, 536.29, 834.25, 466.73, 835.57, 409.32, 823.32, 353.33, 799.63, 299.88, 767.57, 252.6], [691.16, 503.42, 685.09, 520.14, 677.33, 536.27, 669.07, 552.2, 658.45, 567.61, 646.41, 582.22, 633.91, 596.1, 618.12, 608.89, 602.16, 620.88, 586.82, 631.08, 567.43, 641.13, 549.21, 649.59, 530.22, 655.5, 509.37, 662.56, 489.69, 666.26, 465.69, 668.09, 444.74, 668.8, 420.8, 669.85, 398.47, 667.52, 183.82, 523.64, 178.23, 506.43, 172.32, 490.57, 170.84, 472.56, 169.61, 455.29]

[BallX,BallY....],[ZeroX,ZeroY]

Detection will occur only when ball and zero are on screen (wich is not the case in the current version)
I can update my current version and re add the ball position if you want, from the launch to 128 tick records
Screen capture size is 864x864 pixels for information

Thanks !

[Mortimer]

Hi Robotositos!

Sorry my words,but You are on wrong way.Database,ball speed,pivot is doesn't matter.Something 5 another thing very important around
live roulette visual ball strategy.The live roulette working by algorithm,many many times they restarted 8-10 times in a row
same combinations,when focusing for big bet players,or they want more wins him,or the software want  kill his balance.
Complex things.
We developed a visual ball+algo strategy for 4-6 numbers per bet.Max distance for hit (so far) 27 spin.
If you want know exactly the next number hit,it is will fail every time long term,is mission impossible,
because the airball tables,working by magnets.
If you are a programmer maybe we can working together,but you need to know,i DO NOT WANT share here on this forum any details.
Send me mail,if you are interested.
Thanks

[Herby]

ball speed,pivot doesn't matter

This tells clearly Mortimer has no clue of physics.

Physical laws are not negotiable.

A physical model has necessitative (hey I learned a new word) to do with initial positions, velocities and forces.

There is a little chance that Mortimer found a shortcut without knowing physical foundations.

[Herby]

Hi Robotositos,
Pictures to your data below.
Timestamps to the data would be needed.
Cheers

[zeitgeist]

Hi Robotositos!

Sorry my words,but You are on wrong way.Database,ball speed,pivot is doesn't matter.Something 5 another thing very important around
live roulette visual ball strategy.The live roulette working by algorithm,many many times they restarted 8-10 times in a row
same combinations,when focusing for big bet players,or they want more wins him,or the software want  kill his balance.
Complex things.
We developed a visual ball+algo strategy for 4-6 numbers per bet.Max distance for hit (so far) 27 spin.
If you want know exactly the next number hit,it is will fail every time long term,is mission impossible,
because the airball tables,working by magnets.
If you are a programmer maybe we can working together,but you need to know,i DO NOT WANT share here on this forum any details.
Send me mail,if you are interested.
Thanks

Can you send me e-mail? I have a lot of data.

[kabal]

this vb is written in python?

[kabal]

this script u download on github, or you wrote it?
send the link please