In this article, we consider the online parameter estimation problem for overparameterized linear regression models. Many estimation methods require at least the interval excitation condition, which is not always fulfilled for such models, especially in normal operation. To relax the required excitation level, we detect linearly dependent columns in the regressor, obtain the reduced model and estimate parameters in a finite time. On the last step, the parameters of the initial model are recovered. The proposed method efficiency is demonstrated on a simple overparameterized model and a magnetic levitation system.