Coal mine environments are inherently complex and hazardous, often characterized by low illumination, heavy dust, and frequent occlusions that undermine the effectiveness of existing safety monitoring systems. However, millimeter-wave (mmWave) radar–based 3D human reconstruction offers the capability to perceive human posture and motion regardless of lighting conditions and visual obstructions, providing a valuable opportunity to enhance underground personnel monitoring and improve the reliability of safety and rescue operations. Therefore, this paper proposes a mmWave radar–based 3D human reconstruction framework specifically designed for coal mine scenarios. This efficient framework leverages these techniques—including a loss function with angular consistency constraints, batch normalization with cosine annealing learning rate scheduling, and a temporal consistency mechanism—to improve the recognition of human poses and actions in underground environments. On this basis, we further develop an intelligent safety monitoring system for coal mine applications, capable of performing unsafe work behavior analysis with early warnings, personnel positioning, and emergency rescue assistance under disaster conditions. Experimental results demonstrate that the proposed method exhibits strong robustness in tasks such as unsafe behavior analysis in coal mines and reconstruction under dark and humid environments, highlighting the potential of radar-based human perception systems for improving safety and intelligent monitoring in coal mines.