An Image-Based Fluid Surface Pattern Model
This work addresses photogrammetric monitoring for offshore oil drilling platforms, but it is incremental as it focuses on comparing existing dimensionality reduction techniques without presenting new results.
The paper tackles modeling ocean surface dynamics from video for marine condition monitoring, proposing a two-stage approach that reduces dimensionality of video data and applies stochastic modeling to simulate observed flow patterns.
This work aims at generating a model of the ocean surface and its dynamics from one or more video cameras. The idea is to model wave patterns from video as a first step towards a larger system of photogrammetric monitoring of marine conditions for use in offshore oil drilling platforms. The first part of the proposed approach consists in reducing the dimensionality of sensor data made up of the many pixels of each frame of the input video streams. This enables finding a concise number of most relevant parameters to model the temporal dataset, yielding an efficient data-driven model of the evolution of the observed surface. The second part proposes stochastic modeling to better capture the patterns embedded in the data. One can then draw samples from the final model, which are expected to simulate the behavior of previously observed flow, in order to determine conditions that match new observations. In this paper we focus on proposing and discussing the overall approach and on comparing two different techniques for dimensionality reduction in the first stage: principal component analysis and diffusion maps. Work is underway on the second stage of constructing better stochastic models of fluid surface dynamics as proposed here.