New blender-based augmentation method with quantitative evaluation of CNNs for hand gesture recognition

<p>In this study, we extensively analyze and evaluate the performance of recent deep neural networks (DNNs) for hand gesture recognition static gestures in particular. To end, captured an unconstrained dataset with complex appearances, shapes, scales, backgrounds, viewpoints. We then deployed some new trending convolution neuron (CNNs) classification. arrived at three major conclusions: i) DenseNet121 architecture is best rate through almost evaluated red, green, blue (RGB) augmentation datasets. Its outstanding most original works; ii) blender-based help to significantly increase 9% accuracy, compared use a RGB cues; iii) CNNs can achieve impressive results 97% accuracy when training testing datasets come from same lab-based or constrained environment. Their drastically reduced dealing collected environments. In particular, validated CNN on dataset. observed significant reduction only 74.55%. This be improved up 80.59% by strategies such as and/or GAN-based data augmentations obtain acceptable result 83.17%. These findings contribute crucial factors make fruitful recommendations development robust hand-based interface practice</p>