利用影像分割和匹配特征进行无人机影像变化检测

Change Detection Based on Segmentation and Matched Features Points for UAV Images

针对无人机拍摄的影像偏角大、投影差明显的问题,提出一种基于影像分割与匹配特征的无人机影像变化检测方法。该方法基于匹配的特征点和分割的单元,以配准误差为缓冲半径进行相关运算,并提出了双向互相关方法来抑制影像分割不一致对变化检测结果的影响。实验结果表明,该方法提高了无人机影像变化检测的精度,对无人机影像由于大倾角所带来的配准误差问题有较好的容忍度,并削弱了无人机影像的投影差对于变化检测的影响。

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.