Continuous sign language recognition from tracking and shape features using Fuzzy Inference Engine
P. V. V. Kishore, D. Anil Kumar, E. Goutham, M. Manikanta · 2016
Fuzzy classifying continuous sign language videos with simple backgrounds with tracking and shape combined features is the focus of this work. Tracking and capturing hand position vectors is the artwork of horn schunck optical flow algorithm. Active contours extract shape features from sign frames in the video sequence. The two most dominant features of sign language are combined to build sign features. This feature matrix is the training vector for Fuzzy Inference Engine (FIS). The classifier is tested with 50 signs in a video sequence. Ten different signers created 50 signs. Different instances of FIS are tested with different combination of feature vectors. The results are compared with our previous work using no tracking and with discrete sign language database. A word matching score (WMS) gauges the performance of the classifiers. A 92.5% average matching score is reported in this work. A through comparisons for FIS gesture classifier between Discrete Cosine Transform features, Elliptical Fourier descriptor features and the proposed hybrid features for continuous sign language videos show a 40% jump in word matching score.