The location of the center of rotation (COR) of joints is a key parameter in multiple applications of human motion analysis. The aim of this work was to propose a novel real-time estimator of the center of fixed joints using an inertial measurement unit (IMU). Since the distance to this center commonly varies during the joint motion due to soft tissue artifacts (STA), our approach is aimed at adapting to these small variations when the COR is fixed. Our proposal, called ArVEd, to the best of our knowledge, is the first real-time estimator of the IMU-joint center vector based on one IMU. Previous works are off-line and require a complete measurement batch to be solved and most of them are not tested on the real scenario. The algorithm is based on an Extended Kalman Filter (EKF) that provides an adaptive vector to STA motion variations at each time instant, without requiring a pre-processing stage to reduce the level of noise. ArVEd has been tested through different experiments, including synthetic and real data. The synthetic data are obtained from a simulated spherical pendulum whose COR is fixed, considering both a constant and a variable IMU-joint vector, that simulates translational IMU motions due to STA. The results prove that ArVEd is adapted to obtain a vector per sample with an accuracy of 6.8$\pm$3.9 on the synthetic data, that means an error lower than 3.5% of the simulated IMU-joint vector. Its accuracy is also tested on the real scenario estimating the COR of the hip of 5 volunteers using as reference the results from an optical system. In this case, ArVEd gets an average error of 9.5% of the real vector value. In all the experiments, ArVEd outperforms the published results of the reference algorithms.