Human Body Action Recognition based on 3D Feature Fusion over RGB Scenes

Human body action recognition (HBAR) is an important area of research in machine learning and image processing due to its vast range of applications. Similarly, estimating various components of human anatomy from RGB scenes are essential to human action tracking. The present study involves the implementation of human key body point detection for pose estimation. During abstracted silhouettes and vital human points, the proposed system has extracted two main features, such as 3D features and distance features. Once the relevant attributes have been incorporated, executing precise tasks and strategies is critical to achieving maximum features. Therefore, we used the t- SNE-based data refinement technique for optimal feature selection. Finally, for training, a classification model random forest is utilized. The presented system is verified on a recognized benchmark dataset, i.e., the UCF sports database. The experimental settings have revealed that our proposed system has attained better performance and outperformed other present state-of-the-art methods regarding mean recognition rate.