Shape geodesics for robust sign language recognition
Kamal Nasreddine, Abdesslam Benzinou · IET Image Processing · 2019
In this study, a novel method of pattern recognition for static gesture recognition is proposed. To deal with this application, the authors should tackle the problem of low variability among shape classes in the gestures databases. The authors’ method is based on robust registration and shape geodesics in shape space with a preliminary step of pose estimation accelerating the processing time. The different gestures are considered individual points in a non‐linear shape space. Similarity between any two considered shapes can be measured by a distance metric on the shape space. For shape comparison, three distances with and without robust norm are proposed and evaluated for the target application. To cope with low shape variability among classes, the best scheme is to use robust norm in shape matching but not in shape comparison. Experiments are conducted on reference databases used in the literature to evaluate static gesture recognition. These experiments show the outperformance of the proposed scheme compared to state‐of‐the art methods.