Machine Vision for Casino Game Monitoring

João Paulo Maurício

Key information

Authors:

João Paulo Maurício (João Paulo Maurício Pimentel)

Supervisors:

Alexandre José Malheiro Bernardino (Alexandre José Malheiro Bernardino)

Published in

12/13/2010

Abstract

To prevent game losses due to mistakes or irregular strategies, or for statistical data acquisition, the casino game industry is interested in automatic monitoring of the games, recording their state for posterior analysis. A method for detection and recognition of playing cards taking into account partial occlusion and rotation is proposed. This is carried out in two phases: a scene is analyzed to detect rectangles with an original method, fitting the card size, and then each detected rectangle is classified according to the figure present on the corner of the card. The cards are all assumed of a known size and scale, and the image to be analyzed has no perspective or optical deformations, although a calibration phase to take this into account has been developed and tested. Three methods for the card recognition stage are studied and compared, evaluating their performance in terms of robustness to brightness and contrast changes and computation time. A template matching method was shown to work successfully and time efficiently, although showing low robustness to lighting changes, whereas an edge based probabilistic rigid model showed robustness to such changes at the expense of a longer computation time. A probabilistic deformable model presented low success and a very long computation time.