Change Detection in Aerial Images

Abstract

Detection of artificial objects in aerial images of a mainly natural landscape is a relatively new domain of research. The huge amount of data captured from aerial platforms requires efficient tools for analysis of this data. In this project we will apply state of the art and newly developed methods to overcome this major challenge, for various applications. Our project is based on a mathematical tool: Riemann-Laplace Operator.

The data from Elbit will be processed using some flows which come from this operator: Bochner-1D, Bochner-2D, Ricci.  After the understanding of these tools and of the algorithm we want to work with, we will analyze and test different flows on the pictures captured from several areas and try to obtain the best change detection using those flows. Furthermore, one of the points of this project is to verify the efficiency of the diffusion on the data after we applied the different flows. We will explain more in detail about the diffusion and t the different kind of flows in the next paragraph.

Flowchart

Results

                     initial image                                                                                                                 image after filtering

Improvements

• Use the auto-correlation to overcome the changes (weather, lightning, zoom…)
• Or other technique of adjustment: “Local Co-Registration Adjustment for Anomalous Change Detection”, James Theiler and Brendt Wohlberg
• Compute a new registration matrix (different from the one given by Elbit)
• Other flows that can be used to determinate changes in messy urban environments