A robust approach to detect moving object in dynamic background is proposed. The feature descriptions are generated by scale invariant feature transform (SIFT) algorithm. Initial match set is obtained according to nearest neighbor distance ratio(NNDR) strategy, and a robust match set is obtained by using the constraint of symmetry. Random sample consensus(RANSAC) algorithm is applied to distinguish features in moving object region from the background. Based on background compensation, the moving object region is segmented by inter-frame difference and morphology operations. Experimental results on real video sequences from moving cameras demonstrate that the proposed algorithm can obtain a set of robust feature correspondences and detect moving object.