Researchers can benefit from information about watercraft movement throughout the world. One way of obtaining this information is via satellites images.
Abstract
Researchers can benefit from information about watercraft movement throughout the world. One way of obtaining this information is via satellites images. There is a trade-off between spatial resolution and the field of view. In this work, we present an algorithm and its implementation to achieve watercraft detection in MISR images. It uses a set of time-lapse coarse satellite images of a sea-surface area. There is a short time difference between consecutive frames. MISR is satellite that provides a time-lapse set of 9 images within minutes, in resolution of 275×275 square meters per pixel.
The problem (or the background)
Our goal is to detect and characterize the motion of all watercrafts in the field of view. This is a challenge even on clear days, since watercraft are typically much smaller than this pixel footprint.
The solution (or the basic approach):
Detection of watercrafts relies on three features: (1) the watercraft’s typical motion, (2) independence of a watercraft’s motion from its surroundings, and (3) the intensity of the pixel that contains the watercraft. The area of a watercraft is smaller than the area that is covered by one pixel. However, a moving watercraft creates a large, bright wake. A wake typically increases the intensity of its corresponding pixel. The value of such a pixel thus is a local spatial maximum. To detect a moving watercraft we use the fact that the local maximum moves from one image to the next in an orderly manner (see algorithm below). Another difficulty in detecting a watercraft is obstruction by clouds, which leads to false negatives. Another problem is the false positives. Local maxima in moving clouds might lead to false positives. In the chapter “Further suggestions” we suggest ways for elimination of candidates that move similarly to their surroundings. False positives can also be a result of random local maxima that move similarly to watercrafts. Another feature that helps reduce false positives is the radiance: for watercraft wake the radiance is typically significantly less bright than that of a cloud, and more bright than that of random local maxima in sea area that move similarly to watercrafts.
First we identify and list the local maxima in the first image. For each local maximum, we search in the following image for local maxima that are in its proximity. The two corresponding maxima form a pair. Each pair indicates the location of a local maximum expected in the third image. If indeed that image has a spatially consistent maximum point, we have a triplet. Each triplet represents a maximum point moving in constant velocity across three images. We thus have a list of triplets, each being a candidate. The watercraft candidacy list is narrowed further in the next (4th) image etc. leading to a list of consistent quartets, and so forth. Points that move consistently across 8 out of 9 images (from the first image to the last or vice versa) can be indicated as watercrafts.
Jet Propulsion Laboratory, Caltech