A Fast and Accurate Initialization Method for Mocap-Inertial Navigation System

Markerless motion capture (Mocap) has the capability to provide precise positioning, and the integration of Mocap with the inertial measurement unit (IMU) holds significant promise for enhancing 6-DoF pose accuracy and stability in augmented reality (AR). To address the challenge of effectively initializing the system states in the Mocap-inertial navigation system (MINS) within a short timeframe, this article proposes a fast and accurate initialization method. First, to rapidly determine the initial rotation matrix, a novel Mocap-based observation vector construction scheme is presented, which establishes the connection between the global frame and the IMU frame and mitigates the effect of accumulated IMU biases. Second, a joint optimization scheme based on a new state-space description is proposed to estimate the pose, velocity, and system error terms. These estimates are then employed to reduce the observation vector errors and improve the accuracy of initialization. Experimental tests on both the simulation and real-world experiment show that the proposed method exhibits faster convergence of the rotation error and superior estimation accuracy of the 6-DoF pose compared to existing methods for MINS.