Generating Character Animation using Phase Factor Neural Network with Convolution Block Attention Module based DenseNet

In recent years, the endless advancement of technology in 3D animation become popular among peoples due to its benefits of being realistic and stereoscopic. Thus, 3D animation has progressively become the typical technology of commercial and gaming animation. However, an action 3D animated character generation is challenging due to inaccurate estimation of human pose recognition is poor which is performed by the existing Deep Learning (DL) approaches. To overcome this limitation, a Phase Factor Neural Network with Convolutional Block Attention Module based DenseNet (PCBAM-DenseNet) is proposed for correct estimation and recognition of human poses to enhance Three-Dimensional (3D) animation character generation. The DenseNet ability to reuse the features obtained from previous layers with selective focus on important spatial regions and feature channels by PCBAM, lead to improve accurate results for pose recognition. The combination of PCBAM -DenseNet architecture is more suitable for handling complex, noisy, and occluded 3D human pose images efficiently and enhance the estimation precisely. Experimental results of the proposed method achieved accuracy of 91.53% which is greater when compared to the existing approaches of Convolutional Neural Network (CNN).