Teleoperation serves as a vital means of interaction between humans and robots, aiming to enable robots to move in accordance with human intentions. An effective teleoperation system can facilitate seamless collaboration and communication between humans and robots, enhancing their cooperative capabilities. This paper presents a motion-capture-based upper-body teleoperation system for humanoid robots, called TeleMotion, which consists of two key modules. The first module is an inertial sensor-based motion capture subsystem that accurately tracks human motion while remaining unaffected by environmental factors such as lighting and occlusion. The second module is a learnable temporal neural network inverse kinematics algorithm (TNIK) that fully leverages the relationship between historical human motion data and robotic joint angles. This allows for the rapid and precise mapping of human motion to humanoid motion. By integrating these two modules, TeleMotion enables a highly natural and intuitive interaction method for real-time teleoperation of humanoid robot.