Abstract: In this study, a new improved real time Adaptive Constrained Quaternion Extended Kalman Filter (ACQEKF) algorithm is proposed. It is employed to estimate the quaternion and bias states of a constrained nonlinear system perturbed by noise using noisy measurements. The values of the process and measurement noise covariances Q and R, respectively are unknown or partially known, their biased initializations result in the degradation or divergence of the quaternion Extended Kalman Filter (EKF) performance. This study proposes a new method to improve the EKF performance against the covariances uncertainty. Unlike, the previous methods, this method adopts the idea of the recursive estimation of the EKF to propose two tunable recursive updating rules for Q and R, respectively designed based on the filter innovations. As for the quaternion constraint, it is projected onto the EKF gain derivation. The proposed ACQEKF proved itself to have a dramatic improved performance over the conventional EKF, the estimates are more accurate have less noise and more stable.