Improvement of the feature tracking and patter matching algorithm for sea ice motion retrieval from SAR and optical imagery

Abstract

This study presents an improved, versatile, and efficient algorithm based on the Oriented FAST and Rotated BRIEF (ORB) combined with the maximum cross-correlation (MCC) (ORB-MCC) for extracting sea ice motion (SIM) vectors. Quadtree ORB (Q-ORB) extracts more uniform feature points than ORB (uniformity is 3 times higher) and eliminates the concentration of ORB-extracted feature points on ice ridges, leads and coastlines, thereby providing excellent initial conditions for MCC calculations. In addition, a geographic grid-based matching (GGM) algorithm is developed to replace the brute-force matching algorithm (BFM). GGM is 8–10 times more efficient for matching feature points than BFM, thereby increasing the computational efficiency of extracting SIM vectors. Moreover, a locally consistent (LC) flow field filtering process is incorporated to facilitate the filtering of the SIM field. Combining cross-correlation-coefficient-threshold (CCCT)-based and LC filtering processes eliminates erroneous vectors more efficiently than using a CCCT-based filtering process alone. The improved algorithm, named Q-ORB-MCC, is used to extract SIM vectors from imagery acquired by the Sentinel-1 Synthetic-Aperture Radar (SAR), Envisat Advanced SAR (ASAR), Phased Array type L-band SAR-2 (PALSAR-2) onboard the Advanced Land Observing Satellite-2 (ALOS-2), and Moderate Resolution Imaging Spectroradiometer (MODIS). An analysis of the accuracy and effectiveness of the extracted SIM vectors shows that Q-ORB-MCC extracted SIM vectors from Sentinel-1, ASAR, and MODIS images with 4%, 253%, and 62% higher accuracy than ORB-MCC, respectively. Meanwhile Q-ORB-MCC could obtain more SIM vectors from Sentinel-1 and ASAR images.

Keywords

• Sea ice motion
• Feature tracking
• Maximum cross-correlation
• Q-ORB
• Geographic grid-based matching
• Locally consistent filtering