The statistical characterization of the sea for the segmentation of maritime images

Abstract

This paper describes part of an imaging system to indicate possible collision situations between maritime vessels. History shows us that any collision involving a marine craft can have a devastating effect both on human life and the environment. Presented here is a method for the statistical characterization of the sea present in an image and the subsequent segmentation of the image into two regions, sea and object. The characterization process uses statistical measures on histogram data collectedfrom various regions of the image. This leads to a characteristic grey-level range that represents the 'sea ' . The grey-level statistics are then used to segment the image into regions that contain sea. The methods introduced here have been shown to work well with images containing difering sized objects with differing sea conditions and levels of illumination.

Water detection through spatio-temporal invariant descriptors

“…Water detection has been investigated in specialized systems, including autoonomous driving systems [11,31,32], maritime environments [12], and using flying robots [13,14]. Although these algorithms might provide a suitable solution in their restricted environment, none of the mentioned works are able to generalize to fully automatic water detection using minimally constrained video material.…”

“…In all works, additional sensors are used to help the detection problem. In both maritime settings and in works using flying robots, similar non-generalizable assumptions have been made, whether it is assuming that water is within a specific part of the frame [12], requires a manual pre-processing step to identify sky regions [13], or uses auxiliary sensors [13,14]. The works do therefore not generalize to water detection with minimal camera assumptions and without additional sensors, rendering them impractical for the problem of this work.…”

“…Furthermore, the focus is generally on classification or segmentation, but not on the joint problem as posed here. On the other hand, water detection in specialized settings, such as autonomous driving [11] and in maritime settings [12], either make non-generalizable restrictions on the movement and orientation of cameras [11] or use auxiliary data sources in their measurements [12,13,14]. To address the limitations of related work with respect to water detection specifically, this work provides an investigation into the temporal and spatial behaviour of water scenes.…”

On the Segmentation and Classification of Water in Videos

“…Traditionally, automatic water recognition is studied from two perspectives, namely as part of larger recognition tasks such as material recognition (Sharan et al, 2013;Hu et al, 2011;Varma and Zisserman, 2005) or dynamic texture recognition (Chan and Vasconcelos, 2008;Fazekas and Chetverikov, 2007;Saisan et al, 2001;Zhao and Pietikäinen, 2007), and in specialized and restricted environments, including autonomous driving systems (Rankin and Matthies, 2006) and maritime settings (Smith et al, 2003). Most current works in material and texture recognition are based on the hypothesis that target classes can be discriminated using distributions of local image features, global motion statistics, or learned ARMA models.…”

Real-Time Detection of Small Surface Objects Using Weather Effects

“…To avoid collisions of surface vessels, Sanderson et al have proposed some maritime targets identification algorithms [16,17], using statistical characteristics of the sea and motions of the targets. Sullivan et al [18] use an optimal trade-off MACH filter to detect vessels from maritime surveillance videos.…”