SELFIE SIGN LANGUAGE RECOGNITION WITH MULTIPLE FEATURES ON ADABOOST MULTILABEL MULTICLASS CLASSIFIER

The objective is to take sign language recognition towards real time mobile implementation as a communication link between hearing impaired and normal people. Selfie mode continuous sign language video is the capture method used in this work, where a hearing-impaired person can operate the SLR mobile application independently. To decrease the computations per frame, the algorithms are developed for mobile platforms. The operating algorithm consists of key frame extraction, face detection, hand search space identification, head-hand portion extraction, fuzzy hand -head shape segmentation for multiple features. Hand -head configuration, phraseology, shape signature and orientation features are fused to form a dataset. Training with multiple features on the Adaboost Multilabel Multiclass learning algorithm makes the sign classifier faster. Results show excellent recognition rates and faster recall times when compared to state of the art backpropagation learning algorithm with single and multiple deep layers. The system is tested multiple times with 10 different signers with constant video backgrounds for 10 continuous Indian sign language sentences formed from 282 words.