Water detection through spatio-temporal invariant descriptors

Abstract

In this work, we aim to segment and detect water in videos. Water detection is beneficial for appllications such as video search, outdoor surveillance, and systems such as unmanned ground vehicles and unmanned aerial vehicles. The specific problem, however, is less discussed compared to general texture recognition. Here, we analyze several motion properties of water. First, we describe a video pre-processing step, to increase invariance against water reflections and water colours. Second, we investigate the temporal and spatial properties of water and derive corresponding local descriptors. The descriptors are used to locally classify the presence of water and a binary water detection mask is generated through spatio-temporal Markov Random Field regularization of the local classifications. Third, we introduce the Video Water Database, containing several hours of water and non-water videos, to validate our algorithm. Experimental evaluation on the Video Water Database and the DynTex database indicates the effectiveness of the proposed algorithm, outperforming multiple algorithms for dynamic texture recognition and material recognition by ca. 5% and 15% respectively.

People and Vehicles in Danger - A Fire and Flood Detection System in Social Media

“…Unfortunately, BowFire dataset contain pictures only for fire events, so we led to the temporary solution to use VideoWaterDB [25] which contain different video shots from water scenarios, so as to simulate a flood scenario event. Another unfortunate fact in our localization experiments is that that the groundtruth masks that BowFire provides contains only fire pixels, no matter if flood or large water bodies exists as well, which usually happen in real case scenarios (i.e.…”

“…Image localization evaluation took place on BowFire [11] and VideoWaterDB [25] datasets for fire and flood segmentation respectively. Comparisons regarding the fire segmentation results took place on BowFire by computing recall and precision metrics and are depicted in Table II.…”

Momental directional patterns for dynamic texture recognition

“…For instance, with (P, R) = (8, 1) of riu2 mapping, EMDP obtains 1.5% better than that of MDP (see Tables 4 and 9). In the setting chosen for comparison with the state of the art i.e., (8,1), (16,2), (24,3) riu2 , EMDP outperforms about 0.3% compared to MDP (99.43%) on DynTex35. Particularly, it gains 93.94% rate of recognition on the complicated dataset, Gamma, about 2% higher than MDP's.…”

“…Efforts of analysis to make them more "understandable" are crucial for important tasks of recognition, segmentation, synthesis, and indexing for retrieval. Those are primary keys in a large range of applications in computer vision, such as visual surveillance of traffic scenes, crowded people [2], human interaction [3,4,5,6], detecting objects and events [7,8], tracking motion objects [9], etc. The major challenges in DT analysis are due to the wide range of appearances and non-directional motions of DTs.…”

Water Region Extraction in Thermal and RGB Sequences Using Spatiotemporally-Oriented Energy Features

“…These methods make use of color, texture and spatiotemporal statistics of images to form the feature vector. In [18], after a pre-processing step focusing on increasing the invariance against water reflections and colors, spatiotemporal descriptors are used to locally classify the presence of water. This work generates a water detection mask through spatiotemporal Markov Random Field regularization of the local classifications.…”