In this paper we address the problem of joint channel and frequency offset estimation and tracking in multiple-input multiple-output (MIMO) OFDM systems for mobile users. The proposed method stems from extended Kalman filtering and is suitable for time-frequency-space selective channels. Separate offset for each MIMO channel branch is considered because of the mobility and rich scattering. The channel taps and the frequency offsets are estimated in time-domain while the equalization is performed in frequency domain. Simulation results demonstrate that the proposed method tracks time-varying channels and frequency offsets with high fidelity. Realistic channel models are used in mobile scenarios. The proposed time-domain approach has improved performance and robustness in comparison to purely frequency domain processing. Computational complexity is lower as well.