High-precision motion control systems, for instance deployed for micro-and nano-positioning, often use the smart-material based actuators such as piezoelectric and magnetostrictive stages. Those exhibit inherent hysteresis nonlinearities which are challenging to compensate without precise hysteresis modeling. Even if a suitable hysteresis modeling approach is available, its parameter identification, correspondingly adaptation, at normal operating conditions constitute an essential task for the overall control design. This paper uses the direct recursive identification method for the Preisach hysteresis model and describes the fast parallel-computing discrete-time algorithm for an adaptive hysteresis filter. The exponential convergence of errors is ensured up to a residual level proportional to the power of the measurement noise. We provide an explicit discrete-time formulation of the hysteresis filter and its online parameter adaptation. In addition to the theoretical analysis, we demonstrate the expected convergence of the estimation errors by using experimental data from a standard open-loop controlled piezoelectric actuator stage.