IDF-Sign: Addressing Inconsistent Depth Features for Dynamic Sign Word Recognition

Inconsistent hand and body features pose barriers to sign language recognition translation leading unsatisfactory models. Existing models are built up on the spatial-temporal depth Sp features. Finding suitable expert for model is challenging especially dynamic words because many inconsistent exist across motions shapes. In this article, we propose IDF-Sign: an efficient consistent from a multivariate pairwise consistency feature ranking (PairCFR) approach. The temporal obtained by computing 3D position vector of skeletal joint coordinates, while spatial were taking every ten coordinates in video frames averaging it doing so until end frames. PairCFR was used rank select best at different thresholds. We employed threshold selection compute mid-point value each ranked according its weight. receiver operating characteristics (ROC) scheme identify relationship between sensitive parameters features, values utilized as modeling inputs. To verify IDF-Sign, design real-life experiment with leap motion sensor (LMS) consisting signers total ninety words. LMS provides videos, since videos too dense treat directly, read comma-separated files real time. Extensive IDF-Sign evaluations using machine learning ASL, GSL, DSG, ASL-similar datasets prove Optimized Forest achieved average performance 95%, 78%, 65.07%, 95.33% top-1, respectively.