Speaker
Description
As we begin to acquire high-resolution video from space, the potential for new and disruptive satellite-based technologies grows. Real-time airport and road traffic monitoring, increased maritime domain awareness (MDA) through vessel detection and tracking, and timely warning systems for natural disasters are some of the possible avenues being explored. Even in the absence of object motion in the acquired videos, the temporal redundancy could be used to denoise the individual frames, increase the effective resolution through super-resolution restoration, and create 3D products from a single-orbit video.
The successful implementation of most of the above technologies would require the development of artificial intelligence solutions that can be broadly clustered into two classes: (a) data reduction and (b) data mining architectures. The current OVERPaSS project, from UK Centre for Earth Observation Instrumentation is exploring both directions, using data from the recent Carbonite-2 satellite, and is aiming to develop fully functional prototypes that would form the basis of near-future satellite software. In this work we are going to present results on both data reduction and data mining prototypes, demonstrating the potential for this type of on-board software.
Regarding the data reduction, our main focus lies on the fast and accurate detection of clouds in satellite video. Cloud detection (alongside video compression) is the primary strategy needed in order to significantly reduce the data volume that would be downlinked to the ground. By masking out sections of data, or even discarding whole video streams in the case that the surface is completely obscured, we can dramatically reduce the total data throughput of the spacecraft. The cloud detection pipeline, presented in a recent ESA workshop, has been updated and carefully finetuned, leading to a software that performs real-time cloud detection, achieving median estimation error <1% alongside high-quality cloud masks. Several examples of the detection results are going to be presented, while the main conclusions reached will be discussed in detail.
A novel data processing feature to be presented in the workshop will be related to the on-board use of artificial intelligence to extract semantic information from high-resolution satellite videos. Data mining using high-resolution satellite video has a broad range of potential applications. In order to narrow down the application range, we will report on data mining results in two distinct use cases: (a) airport traffic monitoring, including aeroplane detection and temporal tracking, as well as airport runway detection and (b) vessel monitoring, including sea/land classification, vessel detection and vessel speed/trajectory estimation. An initial quantitative evaluation of these results will also be presented.
| Paper submission | Yes |
|---|
