Abstract: Aimed at the reconstruction of 3D motion trajectory in a small range such as human body micro-motion and gesture recognition, the inertial sensor detection method is adopted. The acceleration and angular velocity raw data of the inertial sensor were collected through Zigbee wireless communication, and Kalman filtering was used to preprocess the raw data on the PC side. The fourth-order Runge-Kutta algorithm was used to update the rotation matrix, thus integrating to obtain displacement and compensating to obtain trajectory. Aimed at the problem of inaccurate motion trajectory caused by the cumulative error caused by the integral, a combined zero-acceleration error compensation algorithm was proposed, and the linear interpolation compensation was performed during the motion process to reduce the impact of accumulated errors on detecting motion trajectories within a small range. The experimental results show that the method can effectively improve the accuracy of detecting 3D motion trajectories in a small range.