Ortho Image Mosaicing and Object Identification of UAV Data

Real world applications of UAV imagery are growing at a rate faster than ever. Along with this growth comes the need to process the UAV images and extract useful information from them. This paper illustrates a comprehensive python-based algorithm to stitch multiple images gathered from a single UAV and then perform a landscape scan to identify features and other non-homogeneities in the ortho-mosaiced image. The methodology introduced for image stitching involves key point detection using the SIFT algorithm and key point matching using KNN and RANSAC algorithms. The methodology introduced for object identification involves the computation of intensity changes between blocks of pixels in the horizontal direction (H-Scan). These intensity changes are then sorted and filtered before being mapped to feature types such as house roofs, mud trails, forest cover, etc. depending on the image being analyzed. The results indicate that the algorithm can extract meaningful information such as the location and intensity of features from the ortho-mosaiced image. The computational power required to implement this algorithm is extremely minimal, making it a good preliminary algorithm to use for mosaicing and analyzing a set of overlapping UAV images.