Akademska digitalna zbirka SLovenije - logo
E-viri
Celotno besedilo
  • S, Srithar; L, Arokia Jesu Prabhu; Elangovan, Vetrimani; M, Naveen Kumar; S, Alagumuthukrishnan

    2021 Smart Technologies, Communication and Robotics (STCR), 2021-Oct.-9
    Conference Proceeding

    The deaf and dumb peoples mostly communicate with normal people through the standard nonverbal American gesture. But the sign languages are difficult to recognize and understandable by the common people. The existing computerized sign recognition methodologies suffer from the sign object classification. The false-positive test result reduces the prediction accuracy or increases the training overhead. The proposed approach named Improved Haar Feature (IHF) locate the skeleton of the hand pose and recognize the sign by the Angle Support Vector (ASV). The skeleton is estimated by the series of cyclic connected points of the box model. The object's centroid is estimated, and the cut-vertex is defined through the support vector. The number of vertices and the distance between the point-pin are also considered for region mapping. The number of segregation classes will differ based on the nature of the sign skeleton. The angle of the centroid and the connection points are examined to extract the right feature. The custom gesture is trained using the deep convolutional neural network models. The proposed methodology will observe the hand sign and produce an optimal output in the form of verbal speech or text. The algorithm produces a good prediction accuracy of 96.2 percent.