Fast recognition of postures for a simplified three-fingered artificial hand
- J. Finat, M. Gonzalo-Tasis
- Proceedings IEEE-SMC San Diego
In this paper we develop a hybrid method to establish the correspondence between common triple junctions appearing in two views of the same scene taken by an uncalibrated bicameral heading with known disparity between both cameras and under constant illumination conditions. After a segmentation process, we get a set of edges. Beginning from theses edges, we extract different types of junctions (", T, L or Y). These geometric features are robust even when the noise arising from the real scenes appears. We provide a very fast process of comparison between points placed at a similar depth in real scenes. Next, we apply a labeling relaxation procedure to determine possible homologues between two images. Decision criteria are based on probabilistic methods, allowing us an easy extension under uncertainty conditions or under incomplete information due to partial occlusions or noise.