Rapid and efficient hand gestures recognizer based on classes discriminator wavelet networks

The vision based on hand gesture recognition is one of the key challenges in behavior understanding and computer vision. It offers to machines the possibility of identifying, recognizing and interpreting hand gestures in the aim of controlling certain devices, to monitor certain human activities or interacting with certain human machine interfaces (HMI). This paper aims at proposing a rapid and efficient hand gestures recognizer based on classes discriminator wavelet networks. In this work two main contributions were proposed: firstly, by enhancing previous works using wavelet networks (WN) in the classification field, specifically at the learning process of the latest version of WN classifier (WNC) by creating separator WNs discriminating classes ( n − 1 WNs to classify n classes) as alternative of constructing a WN corresponding to each training image. This contribution, by minimizing the comparison number between test images WNs and training ones, makes quicker the test phase. Secondly, a new WN architecture based on the cascade notion was proposed, in which the WN is decomposed of a set of stages. The novel architecture aims not only at making recognitions robust and quick but also at rejecting from early stages, as fast as possible, gestures that must not be taken into consideration by the system (spontaneous gestures). To test this work, both proprietary and public hand posture datasets were used. Comparisons with other works are detailed and discussed. Given results showed that the novel hand gesture recognizer performances are comparable to already established ones.