Learning Fast Hand Pose Recognition

Practical real-time hand pose recognition requires a classifier of high accuracy, running in a few millisecond speed. We present a novel classifier architecture, the Discriminative Ferns Ensemble (DFE), for addressing this challenge. The classifier architecture optimizes both classification speed and accuracy when a large training set is available. Speed is obtained using simple binary features and direct indexing into a set of tables, and accuracy by using a large capacity model and careful discriminative optimization. The proposed framework is applied to the problem of hand pose recognition in depth and infrared images, using a very large training set. Both the accuracy and the classification time obtained are considerably superior to relevant competing methods, allowing one to reach accuracy targets with runtime orders of magnitude faster than the competition. We show empirically that using DFE, we can significantly reduce classification time by increasing training sample size for a fixed target accuracy. Finally, scalability to a large number of classes is tested using a synthetically generated data set of 81 classes.