Direction of arrival (DOA) estimation using sensor arrays with strong interferences is an important problem in array signal processing. Although the DOA estimation accuracy of existing methods for on-grid targets is high, the DOA estimation performance of these methods for off-grid targets degrades seriously. In this paper, a new DOA estimation method is proposed. Firstly, a new matrix filter called robust orthogonal matrix filter with nulling (ROMFN) which combines the orthogonal matrix filter with nulling (OMFN) and robust adaptive beamforming using worst-case is used for filtering of off-grid targets. ROMFN is effective for off-grid targets just with a few grid points designed. Besides, the new matrix filter preserves the white Gaussian noise property, avoiding the pretreatment of noise whitening. Secondly, a new DOA estimation method is used to estimate DOAs of off-grid sources with strong interferences based on off-grid sparse Bayesian inference (OGSBI) and ROMFN. Compared to previous methods, simulation results validate the proposed method can achieve better resolution in different grid intervals, different SNRs and different INRs.