UAV(unmanned aerial vehicle) images suffer from big registration and projection errors when UAV images are captured due to unstable rotary wings. In this paper, we propose a new method for change detection using UAV images, that compensates for these sources of error. Our method combines feature points matching and image segmentation. By merging the results of unmatched feature points and low-similarity segmented objects, the changed areas will be detected. By using the value of image registration error as searching buffer radius, mutual cross correlation calculations of the corresponding segmented objects are employed to leverage the impact of inconsistent segmentations on change detection results. Experimental results illustrate that the proposed method outperforms traditional methods as it integrates the context texture and spectral information from segmented objects, which can weaken the impact of image registration and projective errors resulting from the large rotation angle and improve the accuracy of change detection to certain extent.