Detection of hand LBP descriptors

Abstract

Gesture recognition has been presented as an alternative for the implementation of efficient interaction. Particularly machine vision-based applications have advantages in portability over other alternatives. However, algori- thms often require computational intensive training, being difficult to implement in mobile devices. This paper presents a preliminary study to detect poses hand using an algorithm based on local binary patterns, better known by its acronym LBP (Local Binary Patterns). Thus, a heuristic model of hand division in regions differentially weighted, which allows direct classification of gestures using a similarity measure, is presented. Weights and distribution regions were evaluated according to their accuracy in the classification of each hand pose. Those tests were performed on a set of images captured in controlled conditions corresponding to five different poses. The proposed algorithm with the scheme of weighted regions shows a good ability to discriminate and presents a viable alternative for future applications.