Pipeline Image Stitching Method Based on Improved SIFT Algorithm

Detection of defects such as cracking, corrosion and wear is critical during the service life of oil and gas pipelines. Ultrasonic, magnetic leakage and visual imaging techniques are commonly used in the industry for evaluation. This group has developed an array side-view imaging inspection device that acquires video images of the inside of the pipeline through four cameras. Due to the limited field of view (150°) of a single lens, a panoramic image must be acquired with the help of image stitching techniques. In this paper, we propose an improved image stitching method with SIFT algorithm, which extracts feature points through scale space construction and feature descriptors, combines Euclidean distance matching with RANSAC algorithm to screen the mismatch points, and utilizes weighted average fusion to eliminate artifacts. The experiment compared the improved SIFT algorithm in this paper with the SIFT algorithm. The results show that the improved algorithm increases the average number of feature point detections by 78.64%, and performs better in aspects such as the image stitching quality evaluation indicators SSIM and NIQE, providing a new idea for the quantitative measurement of pipeline defects.