Single-camera human tracking struggles with complex scenarios with occlusions. Additionally, connecting multiple cameras to a central PC hinders real-time capabilities and scalability . We propose a modular multi-camera human pose estimation system powered by AI vision and edge computing. We developed an edge module for camera image acquisition, 2D human pose estimation, and publishing key pose data. One of these modules is comprised of a Jetson Orin Nano and an RGB camera. A central PC merges pose data from multiple modules to reconstruct 3D human poses. By placing 4 modules around a robotic cell we can achieve near real-time performance.