Key points are not available for this paper at this time.
Video image matching to find tie points is one of important steps for merging dynamic, narrow field-of-view aerial video into a mosaic orthoimage. Because of its high data sampling rates and inherent characteristics of inconsistent and unstable flying of low-cost unmanned aerial vehicle (UAV), automatically matching of video data is still a big challenge and ongoing effort. This paper presents a self-adaptive image matching technique to automatically and effectively seek the conjugate points. The basic steps involves: (1) automatically extracting frame from video stream at real-time according to any given location or time slot, in order to save the post-processing time; (2) self-adaptively adjusting parallax of two neighbor frames to predict conjugate points to reduce searching space; and (3) the accuracy of image matching result is self-estimated using the technique of two-view co-planar geometry constructed by conjugated points and baseline. Based on the estimated accuracy value, the iteration for increasing or decreasing sample time slice to obtain new image pair is determined. The experimental results demonstrated that this proposed method can achieve high automation and real-time processing purpose for response to time-critical disaster applications.
Wu et al. (Sat,) studied this question.