AdaBoost in basketball player identification

Video materials contain huge amount of information. Their storage in databases and analysis by various algorithms represents an area that constantly develops. This paper presents the process of analysis of basketball games by AdaBoost algorithm. This algorithm is mainly used for face recognition and body parts recognition. It consists of a linear combination of weak classifiers. In this paper, stumps were used as weak classifiers. The aim of this research is to assess the accuracy of this algorithm when applied in players' identification at basketball games. Capabilities of AdaBoost were examined when applied to video footage from single moving camera, and when these footages were not previously treated by any other algorithm. The first training was performed by entire images of basketball players, whereas the second training was performed by using the images of the head and torso. By applying the algorithm to the given set of images that include head and torso, the algorithm obtained an accuracy of 70.5%. Experimental results have shown that training on the set of entire body images was not possible due to large amount of background that goes into the training, and which represents noise in training process. This accuracy could be increased by applying filters that would remove background from images and leave just basketball players. By applying those filters, the amount of noise in the training data would be significantly reduced.