Motion capture provides the best interface to the understanding of trajectory planning and formations in the central nervous system, which has enormously benefited the research in interactive game and learning, animation, film special effects, health care, and navigation. The existing human motion capture techniques, which use structured multiple high-resolution cameras in a dedicated studio, are complicated and expensive. With the rapid development of microsensors on chip, a ubiquitous real-time low-cost human motion capture system that uses wearable microsensors becomes possible. Because of the agility in movement, upper limb motion estimation has been regarded as the most difficult problem in human motion capture. In this paper, we take the upper limb as our research subject and present a novel upper limb motion estimation algorithm by hierarchical fusion of sensor data and human skeleton constraints. Our method represents the orientations of upper limb segments in quaternion, which is computationally effective and can avoid the singularity problem. To address the nonlinear human body segment motion, a particle filter is proposed to fuse inertial and magnetic sensor data. To compensate for the drift, which is the most challenging issue in motion estimation using inertial sensors, we present a novel solution by modeling the geometrical constraint in elbow joint and fuse the constraint in the framework of particle filter to revise the sensor fusion results and improve the estimation accuracy. The experimental results have shown that the proposed algorithm can provide accurate results compared to the ground truth.