H
player
=
P
N
i=0
Bet
i
∗ ExpectedHold
i
T otalAmountBet
(3)
H
flatbet
=
AverageBet ∗
P
N
i=0
ExpectedHold
i
T otalAmountBet
(4)
Fig. 7. The gap between these two lines isolates the advantage the player
created by tracking the count and betting accordingly. In this particular case,
the card counting and bet variation shifts the advantage for the house to a
slight advantage for the player.
VIII. CONCLUSION
In this work, we presented a Mask R-CNN based approach
for a new domain of assessing the player’s worth as they
play a game of blackjack. Our method, DeepGamble, takes
images from two viewpoints - chipboard and overhead, to
predict average bet, game outcome, blackjack skill level and
the likelihood of card counting in real time. The proposed
method can easily adapt to different blackjack tables with
different payout rules. Minimal fine-tuning might be needed
as we change scales and perspectives. Extensive experiments
on multiple hands (∼ 150) demonstrated the efficiency and
effectiveness of the proposed method over the state-of-the-art.
We do understand that as dealing style varies from dealer
to dealer, we may have a lot of occlusion in the frames due to
dealer’s hand, which currently our model is able to discard as
fewer feature maps in Mask R-CNN fire on hands of different
sizes. In future work, we want to directly embed the occlusion
of hand as a weighting layer into Mask R-CNN. It will produce
different weights to combine feature map outputs at every pixel
depending on the occlusion.
ACKNOWLEDGMENT
The authors would like to thank Arun Shastri, Rasvan
Dirlea, Mike Francis, Akshat Rajvanshi, Manoj Bheemineni,
Brendan Riley, Geoff Cohn, Jayendu Sharma, Thompson
Nguyen and others who contributed, supported, guided and
collaborated with us during the development and deployment
of our system.
REFERENCES
[1] W. Cooper and K. M. Dawson-Howe, “Automatic blackjack monitor-
ing,” in Proceedings of Irish Machine Vision and Image Processing
Conference (IMVIP 2004). Dublin, Ireland: Association for Computing
Machinery, 2004, pp. 248–254.
[2] S. Auberger, “Blackjack-simulator with omega ii card counting,” 2018.
[Online]. Available: https://github.com/seblau/BlackJack-Simulator
[3] S. Michael, “Buster blackjack side bet house advantage study,” https:
//wizardofodds.com/games/blackjack/side-bets/buster-blackjack/, 2016.
[4] K. Hecht and L. Storch, “Sequenced antenna array for determining
where gaming chips with embedded rfid tags are located on a blackjack,
poker or other gaming table and for myriad other rfid applications,” U.S.
Patent: US20070035399A1, 2005.
[5] H.-S. Yang and C. S. Pyo, “System and method for managing casino
chip using rfid,” U.S. Patent: US8591320B2, 2012.
[6] A. Zaworka and S. Scherer, “Machine vision driven real-time blackjack
analysis,” in Proceedings of 24th workshop of the AAPR, 2000.
[7] C. Zheng and R. Green, “Playing card recognition using rotational
invariant template matching,” in Proceedings of Image and Vision
Computing New Zealand, 2007.
[8] J. Ponce, M. Hebert, C. Schmid, and A. Zisserman, Toward Category-
Level Object Recognition. Springer, 2006, vol. 4170.
[9] K. Mikolajczyk and C. Schmid, “A performance evaluation of local
descriptors,” IEEE Transactions on Pattern Analysis and Machine Intel-
ligence, vol. 27(10), pp. 1615–1630, 2005.
[10] D. G. Lowe, “Distinctive image features from scale-invariant keypoints,”
International Journal of Computer Vision, vol. 60, pp. 91–110, 2004.
[11] “Tangam systems,” http://tangamgaming.com/.
[12] T. Daniel, “Who’s holding the aces now?” http://blackjackuniverse.com/
pages/news/mindplay.html, 2003.
[13] S. E. Inc, “Casino table monitoring/tracking system,” Canada Patent:
CA2463254C, 2002.
[14] S. G. Inc, “Method and apparatus for using upstream communication in
a card shuffler,” U.S. Patent: US10226687B2, 2016.
[15] I. G. T. SHFL Enterteiment Inc, “Gambling chip recognition system,”
U.S. Patent: US5781647A, 1997.
[16] VizExplorer, https://www.vizexplorer.com/.
[17] A. Labs, https://www.chipvue.com/.
[18] K. He, G. Gkioxari, P. Doll
´
ar, and R. B. Girshick, “Mask r-cnn,” 2017
IEEE International Conference on Computer Vision (ICCV), pp. 2980–
2988, 2017.
[19] T.-Y. Lin, M. Maire, S. J. Belongie, B. Lubomir D. title=Microsoft
COCO: Common Objects in Context, R. B. Girshick, J. Hays, P. Perona,
D. Ramanan, P. Doll
´
ar, and C. L. Zitnick, in ECCV, 2014.
[20] W. Abdulla, “Mask r-cnn for object detection and instance segmentation
on keras and tensorflow,” https://github.com/matterport/Mask
RCNN,
2017.
[21] F. Chollet et al., “Keras,” https://keras.io, 2015.
[22] M. Abadi, A. Agarwal, P. Barham, E. Brevdo, Z. Chen, C. Citro, G. S.
Corrado, A. Davis, J. Dean, M. Devin, S. Ghemawat, I. Goodfellow,
A. Harp, G. Irving, M. Isard, Y. Jia, R. Jozefowicz, L. Kaiser,
M. Kudlur, J. Levenberg, D. Man
´
e, R. Monga, S. Moore, D. Murray,
C. Olah, M. Schuster, J. Shlens, B. Steiner, I. Sutskever, K. Talwar,
P. Tucker, V. Vanhoucke, V. Vasudevan, F. Vi
´
egas, O. Vinyals,
P. Warden, M. Wattenberg, M. Wicke, Y. Yu, and X. Zheng,
“TensorFlow: Large-scale machine learning on heterogeneous systems,”
2015, software available from tensorflow.org. [Online]. Available:
http://tensorflow.org/
[23] S. Chetlur, C. Woolley, P. Vandermersch, J. Cohen, J. Tran, B. Catanzaro,
and E. Shelhamer, “cudnn: Efficient primitives for deep learning,” CoRR,
vol. abs/1410.0759, 2014.
[24] J. Alexander, “Imgaug,” https://github.com/aleju/imgaug, 2018.
[25] T. Chen and C. Guestrin, “Xgboost: A scalable tree boosting system,”
in KDD, 2016.
[26] B. C. Russell, A. Torralba, K. P. Murphy, and W. T. Freeman, “Labelme:
A database and web-based tool for image annotation,” International
Journal of Computer Vision, vol. 77, pp. 157–173, 2007.
[27] K. Zutis and J. Hoey, “Who’s counting? real-time blackjack monitoring
for card counting detection,” 10 2009, pp. 354–363.
[28] S. Michael, “Blackjack card counting strategy, high low,” https://
wizardofodds.com/games/blackjack/card-counting/high-low/, 2008.