Multi-temporal Registration of Environmental Imagery Using Affine Invariant Convolutional Features

• Published in: Image and Video Technology Vol. 11854; pp. 269 - 280
• Main Authors: Khan, Asim, Ulhaq, Anwaar, Robinson, Randall W.
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2019; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Deep learning; Descriptors; Image registration; Repeat photography
• ISBN: 3030348784; 9783030348786
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-34879-3_21

Repeat photography is a practice of collecting multiple images of the same subject at the same location but at different timestamps for comparative analysis. The visualisation of such imagery can provide a valuable insight for continuous monitoring and change detection. In Victoria, Australia, citizen science and environmental monitoring are integrated through the visitor-based repeat photography of national parks and coastal areas. Repeat photography, however, poses enormous challenges for automated data analysis and visualisation due to variations in viewpoints, scales, luminosity and camera attributes. To address these challenges brought by data variability, this paper introduces a robust multi-temporal image registration approach based on affine invariance and convolutional neural network architecture. Our experimental evaluation on a large repeat photography dataset validates the role of multi-temporal image registration for better visualisation of environmental monitoring imagery. Our research will establish a baseline for the broad area of multi-temporal analysis.