In the fields of anti-terrorism and emergency rescue, human pose sensing under non-line-of-sight (NLOS) conditions plays a critical role in enabling informed decision-making and effective response. Current research predominantly focuses on line-of-sight (LOS) scenarios but pays limited attention to NLOS conditions, and even these NLOS studies are confined to single-person pose detection, significantly restricting real-world multi-person applications. In this paper, we propose the first method that enables adaptive multi-person pose sensing in NLOS environments. Our approach reconstructs coarse 3D features from transients and refines these features through a neural network. The refined features are then utilized to predict a 3D body center heatmap and a mesh parameter map. Finally, using 3D body center-guided parameter sampling, the method enables adaptive multi-person 3D mesh regression. Comprehensive experiments on both simulated data and real-world data acquired using our self-built NLOS system demonstrate that our method effectively senses human poses in multi-person NLOS scenarios. Notably, the simulation pipeline we developed for constructing the multi-person NLOS dataset and the real-world data captured by our system provide valuable resources for advancing research in this field. We will publicly release our dataset to support the development of new methods and foster progress in NLOS human pose sensing.