Automatic synthesis of classifiers in FGPA

In this paper an approach to automatic synthesis of classifiers in FPGAs based on AdaBoost algorithm is introduced. The various boosting methods are widely used for detection of objects in images. In this paper a novel approach to automatic synthesis of a classifier with respect to an efficient use of computing resources is proposed. The main purpose is to achieve an acceptable trade-off between speed and resources consumption of the designed classifier. The features, used in boosting algorithms, are based on Local Binary Patterns, which are more suitable for hardware implementation then traditionally used Haar-like features. Automatic synthesis of classifiers from trained data sets is also described. In the paper there are also described the results achieved.