With the growing integration of artificial intelligence and IoT technologies in sports training, enhancing training efficiency and accuracy has become increasingly important, especially in martial arts, where complex and rapid movements require precise evaluation. This study proposes ST-LineNet, a novel IoT-based model for real-time 3D pose estimation in martial arts training. ST-LineNet incorporates spatiotemporal transformers to capture temporal dynamics and multi-scale attention mechanisms for spatial feature extraction, achieving high precision and real-time feedback. Experimental results show that on the Human3.6M dataset, ST-LineNet achieved an average MPJPE of 28.3 mm under Protocol #1 and 32.0 mm under Protocol #2. Additionally, on the MPI-INF-3DHP dataset, the model reached a PCK of 98.2%, an AUC of 77.8, and an MPJPE of 29.8 mm, while maintaining a real-time processing speed of over 30 frames per second. These results validate the effectiveness of ST-LineNet for enhancing martial arts training efficiency. This system offers timely, objective feedback for martial arts practitioners, demonstrating the potential of IoT-enhanced sports training solutions.