Human Action Recognition Utilizing Doppler-Enhanced Convolutional 3D Networks

While significant advancements have been made in DL-based human action recognition (HAR), accurately classifying athletes' actions remains challenging, primarily due to the need for comprehensive sports athletes' datasets. Recognizing the limited availability of accessible athlete action datasets, we have proactively taken the initiative to develop two meticulously tailored datasets designed explicitly for sports athletes, subsequently assessing their impact on improving performance. While 3D convolutional neural networks (3DCNN) outperform graph convolutional networks (GCN) in HAR, they demand signif-icant computational resources, especially with large datasets. Our study introduces innovative strategies and a more efficient solution for action recognition, reducing the computational load on the 3DCNN. Therefore, it offers a multifaceted solution for enhancing HAR, which bridges gaps, tackles computational challenges, and significantly advances the accuracy and efficiency of HAR.