IJFANS International Journal of Food and Nutritional Sciences

ISSN PRINT 2319 1775 Online 2320-7876

Spatial-temporal color mappings for recognizing 3D skeletal sign language

Main Article Content

Ch.Raghava Prasad
» doi: 10.48047/IJFANS/11/S6/002

Abstract

As a result of their widespread availability, feature maps do not discriminate between similarly worded signs. When using unnormalized training maps, traditional CNNs struggle to isolate non-discriminatory features from feature maps due to the presence of vanishing gra dients. The loss of information caused by the vanishing gradients in the deeper CNN layers makes it challenging to develop discriminative features for recognition. In order to correctly represent 3D joint motions, we propose a novel color-coded feature map, joint angular velocity maps (JAVMs). We propose a new ResNet architecture, termed connived feature ResNet (CFR), as an alternative to standard convolutional neural networks. In comparison to other ResNet and CNN based architectures utilized for sign classification, this one does not employ dropout in the final layers and achieves the required result in a smaller number of iterations.

Article Details