AZTI-Tecnalia (www.azti.es), a technological center with expertise on marine and food research, works on the study of coastline dynamics with the goal of establishing its relationship with its morphology and usages, given that this relationship is the one controlling, for instance, the flooding of maritime walks, beach configurations, or the suitability of conditions for leisure uses of the coastline. Of special interest is the analysis of the dynamics of beaches where a high degree of urbanization and exposure to the sea, together with a great environmental and economic value, makes them be in a fragile equilibrium.

Spatial (from hundreds of meters till kilometers) and temporal (hours, days) scales characterizing the dynamics of the coast imply the unsuitability of conventional, limited and costly measurement techniques, to study the behavior of these coastal system. Fortunately, image and video-based measuring techniques (commonly known as coastal video-monitoring) allow us nowadays to describe physical process across a wide range of spatial and temporal scales, something that was considered as almost impossible not much time ago.

A coastal video monitoring system consists of a series of cameras installed along the coast that capture images. Products derived from the processing of the obtained images can give direct and precise information that is of interest for the variety of activities that take place in the area, and that depend on the swell, currents and tides (hydrodynamic conditions), as well as of the configuration of the beach, sand dunes, canals and bars (sedimentary elements). But attempting to go one step further, building on top of these tools, in recent years an important advance has been achieved in the capability of making trustworthy forecasts of sea conditions in beaches.

Availability of updated morphology information in the most shallow region (wave breaking area) to simulate process, and hence prediction and prevemtion (water quality, currents, temporals, etc.) is fundamental, since it is this morphology what totally conditions swellage characteristics (height, for instance) and the localization and intensity of currents. Traditionally, topography and bathymetry are performed by means of a variety of techniques that need to be transported by foot or onboard. This greatly limits the accessibility, especially in the low tide waves break area sand stripe for open beaches, apart from the increased cost they imply for a continuous tracking. At the same time, this represents a challenge for the monitorization and continuous update of the beach morphology, rendering hydrodynamic models useless.

The LOREA Project (POCTEFA 2009-2012) set up the launch of local hydrodynamic models, and within the ongoing project PRE2PLA (POCTEFA 2013-2015), tools are being developed in two pilot beaches, Zarautz and Anglet, to integrate into the existing models updated morphological information of the beach, acquired from video monitoring images. The system aims to be a tool capable of providing practical information about the needs from the point of the view of the management of the sandy area and its environment, as well as from the point of view of the different leisure usages that it receives. The main services would be focused on the detection and prediction of potentially dangerous bath areas, flood alarms due to swell, or emergency management.

AZTI-Tecnalia has put to practice and validated a methodology to obtain in a routine basis the topo-bathymetry associated to the most swallow area of the beach, starting from the combination of video-monitoring techniques with numerical models. To that end, it is necessary to analyze and process the obtained images in an automatic and trustable manner. It is therefore essential to implement image pre-processing and improvement methods. In this line, AZTI-Tecnalia has been collaborating within the same Tecnalia Corporation with the Computer Vision area at TRI. This is a team with expertise in image enhancement and pre-processing, as well as automatic image interpretation by means of machine learning techniques. A previous recent result of this collaboration has been another project to develop an automatic system to measure beach occupancy from static cameras. Within the current project, the Computer Vision Group at Tecnalia contributed to the normalization of image characteristics to be able of extracting the energy disipated by the wave, a key parameter for the computation of the hydrodinamic model developed by AZTI-Tecnalia.

Problem Description

AZTI-Tecnalia methodology to obtain bathymetric maps (sea bottom relief) is based on a mathematical equation relating sea depth to the energy released by wave fronts at the breaking moment. For that, by means of cameras, information related to the amount of water foam generated by breaking waves is retrieved from averaged images (TIMEX). This is achieved by means of a set of images of the beach from a single point of view, as can be appreciated below:

Figura 1

After proper orto-rectification in the sand area of these images, they are combined in a single mosaic, from which an analysis of the chromatic components in the wave breaking area is performed. An example of the mosaic corresponding to the previous images can be appreciated below:

Figura 2

Starting from this mosaic, a transformation to a HSV space is computed, and the resulting image is normalized with the average intensity value of the non-breaking region of the water, assuming that this is located on the upper part of the image.

Next, with images obtained from cameras averaged along 20 minutes and orto-rectified, and with the boundary conditions (wave fronts and tides), energy dissipation maps are derived. To do that, we stablish the hypothesis that wave breaking areas are characterized by the averaged image regions of higher intensities (white sea foam).

The total amount of energy of the incident waves is computed in the region of the beach corresponding to the images, along the averaging time (20 minutes), and it is adjusted with a transformation function (Eq. 1) between energy dissipated by breaking waves (total incident) and luminic intensity per surface unit considered.

Figure 3: Equation relating the transformation between wave energy and luminic intensity.

This way, dissipation maps as the one below are obtained.

Figure 4. Rectified image and obtained dissipation map.

Therefore, it is essential for the intensity range within the wave breaking area to be as wide as possible, without exhibiting artificial intensity discontinuities in the breaking region. As can be appreciated in the previous image, in the process of combining the orto-rectified images (image stitching), such artificial discontinuities are being introduced.

AZTI-Tecnalia and the Computer Vision group’s goal in this project is to solve the negative output of the stitching mentioned above. The Computer Vision group proposed an alternative method consisting of two steps:

  • Color correction of the input images.
  • Iterative blending of the images. This consists of:
    • Corner detection on the overlapping area between the images.
    • Blending masks computations, specifically adapted to the overlapping are between images.
    • Effective blending of the images.

Color Correction:

Once the overlapping área between two images has been located, first-order statistics of the image to be corrected are mapped to the statistics of the same are in the other image.

Figura 5

The result of this mapping depends on which image we take as reference. Given the need to attain a maximum dynamic range in the wave breaking area, we take as reference image the one for which the correction produces the lesser quantity of saturated pixel in the blue channel of both images. This amounts to minimizing the loss of information when converting the color, something that is considered as essential for the correct performance of the posterior bathimetry estimation algorithm.

Blending:

In a first step, the four corners corresponding to the overlapping área are detected, as can be appreciated in the above image. These corners are employed to map a rectangular mask containing values linearly distributed in the [0,1] interval.

Figura 6

The computed linear mask is transformed to the domain of intersection of both images as shown below.

Figura 7

This mask, together with its inverse, are the ones that makes possible a linear combination of both images in such a way that discontinuities are avoided in the overlapping area. The next step is to compute a linear combination of both images, with weights given by the computed masks. This process can be appreciated in the image below.

Figura 8

The output of this procedure is shown below.

Figura 9

This process is completed iterating these steps for the rest of the images. Below we can appreciate the improvement provided by the method proposed by the Computer Vision group, with respect to the process suggested initially by AZTI-Tecnalia.

Figura 10

More results are shown next:

Figura 11