Research On POS Correction And Image Stitching Of UAV

Because UAV has many advantages such as light weight, small size, easy to transport, mapping cycle is short, timeliness, low cost, and can reach personnel cannot enter the danger zone, and therefore favored by many fields, especially in the large-scale mapping operations advantage. UAV is not only a sublimation of satellite remote sensing, it is a complementary geodetic instruments. With UAV widely used in surveying and mapping, the image quickly and efficiently stitching and DEM production more and more important. But UAV is also flawed, such as by airflow impact, unstable posture, rotary angle is large, POS accuracy is not high, as the frame size is small so that more number of photographs and other issues, resulting in image matching slow, or even failure. Especially in earthquake relief and the urgent need mapping for field survey, so cannot meet the requirements of image mosaic speedly and a variety of additional product.Based on the characteristics of UAV image, combined with the two main problems encountered in the production- POS accuracy is not high and photograph overlap level or posture angle do not satisfy the requirements of traditional aerial triangulation, the paper focused on these two issues makes a Study and proposes solutions. For the first question- the match fails because of POS problems, it proposes an approach to give initial value of POS line elements and correct. First with SIFT operator on image feature extraction, and then based on the Euclidean distance matching, next based on the principle of image correspondence points equaling to corresponding model point coordinates gives roughly initial value of POS line elements, following with continuous process relative orientation to correct the initial value, calculating the new POS line elements. To assist complete matching and the follow-up work, used it in aerial triangulation software again. In addition for the second question, in a sequence of UAV image do not satisfy the three more degrees overlapping and small angle, etc., according to the traditional aerial triangulation process cannot be completed relatively orientation and image stitching, the article image mosaic based on computer vision stitching, presents with SURF operator to detect the feature point and with SIFT operator extracting descriptors, based on BBF match, use RANSAC eliminate mismatches, using the similarity transformation, affine transformation, projective transformation to establish geometry transformations model respectively, the study found the accuracy of projective transformations highest, more suitable for UAV image mosaic, and use LM local optimization algorithm to optimize the parameters of the model, without POS and camera calibration data, as long as image owning overlap(large angle degree or overlap cannot reach photogrammetric production requirements) can achieve matching and stitching fastly. Finally, large image using POS data to simple georeferencing, and then after resampling, image will have substantially the coordinate information, having far-reaching significance for emergency disaster and field surveying and mapping guidance work.