IMPLEMENTATION OF PANORAMIC IMAGE STITCHING ON DRONE-CAPTURED IMAGE FOR LABORATORY SCALE MAPPING

• Main Author: Imron Catur Anoraga, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/81538
• Institution: Institut Teknologi Bandung

This research was conducted to design an image stitching algorithm to obtain a wide image that accurately represents the actual conditions and to determine the maximum overlapping area that can still be mapped. The images used are aerial images captured by an Unmanned Aerial Vehicle (UAV), specifically a drone. Additionally, this research aims to determine the optimal overlap area in a laboratory-scale mapping scenario. The stitching algorithm is designed using the panoramic image stitching method. Feature extraction is performed using the Oriented FAST and Rotated BRIEF (ORB) method for each color channel and on four parts of the image. Feature matching is carried out using a brute force approach with a maximum image shift constraint. Mismatch removal is done using Random Sample Consensus (RANSAC) by selecting points within a certain radius. Image registration is performed using a homography transformation model. Images are stitched together by creating masking and gradients at the mask edges. To evaluate the efficiency of the developed algorithm, an analysis of processing time and computational load is conducted, shown through algorithm complexity using Big O Notation. Furthermore, to determine the similarity level between the stitched image and the reference image, the Structural SIMilarity index (SSIM) metric is used. The results show that the algorithm can stitch 90 – 140 images in 18 – 35 minutes with a total complexity of O(niIM). Additionally, the SSIM metric test results indicate that the stitched image sufficiently represents the actual conditions with a score of approximately 0.76 to 0.84. Keywords: mapping, UAV, ORB, RANSAC, Big O Notation, SSIM.