Annette Stahl

Variational estimation of experimental fluid flows with physics-based spatio-temporal regularization

We present a variational approach to motion estimation of instationary experimental fluid flows from image sequences. Our approach extends prior work along two directions: (i) the full incompressible Navier–Stokes equation is employed in order to obtain a physically consistent regularization which does not suppress turbulent variations of flow estimates; (ii) regularization along the time axis is employed as well, but formulated in a receding horizon manner in contrast to previous approaches to spatio-temporal regularization. This allows for a recursive on-line (non-batch) implementation of our variational estimation framework. Ground-truth evaluations for simulated turbulent flows demonstrate that due to imposing both physical consistency and temporal coherency, the accuracy of flow estimation compares favourably even with advanced cross-correlation approaches and optical flow approaches based on higher order div–curl regularization.

On-Line variational estimation of dynamical fluid flows with physics-based spatio-temporal regularization

We present a variational approach to motion estimation of instationary fluid flows. Our approach extends prior work along two directions: (i) The full incompressible Navier-Stokes equation is employed in order to obtain a physically consistent regularization which does not suppress turbulent flow variations. (ii) Regularization along the time-axis is employed as well, but formulated in a receding horizon manner contrary to previous approaches to spatio-temporal regularization. This allows for a recursive on-line (non-batch) implementation of our estimation framework. Ground-truth evaluations for simulated turbulent flows demonstrate that due to imposing both physical consistency and temporal coherency, the accuracy of flow estimation compares favourably even with optical flow approaches based on higher-order div-curl regularization.

Splash detection in surveillance videos of offshore fish production plants

Automatic detection of fish welfare related parameters is a very important step in the process of aquaculture production control. Poor handling, and lack of control of the state of the biomass in production plants, may lead to various disease outbreaks, chronic stress and physical trauma, which can influence mortality, which is directly related to profit loss. Automated and objective splash detection provides reliable information about surface activity, which may provide valuable insight into the state of the fish in the cage. In this paper, we propose an algorithm based on Support Vector Machines (SVM), for automatic splash detection in plant surveillance videos, obtained using an unmanned aerial vehicle. We also evaluate the use of Bag-of-Words (BoW) and Vector of Locally Aggregated Descriptors (VLAD) descriptors, for use in splash detection algorithms.

A New Framework for Motion Estimation in Image Sequences Using Optimal Flow Control

Application of tools from optimal flow control to the field of computer vision and image sequence processing, has recently led to a new and promising research direction. We present an approach to image motion estimation that uses an optimal flow control formulation subject to a physical constraint. Motion fields are forced to satisfy appropriate equations of motion. Although the framework presented is flexible with respect to selection of equations of motion, we employ the Burgers equation from fluid mechanics as physical prior knowledge in this study. To solve the resulting time-dependent optimisation problem we introduce an iterative method to uncouple the derived state and adjoint equations. We perform numerical experiments on synthetic and real image sequences and compare our results with other well-known methods to demonstrate performance of the optimal control formulation in determining image motion from video and image sequences. The results indicate improved performance.

Vision based obstacle avoidance and motion tracking for autonomous behaviors in underwater vehicles

Performing reliable underwater localization and maneuvering of Remotely Operated underwater Vehicles (ROVs) and Autonomous Underwater Vehicles (AUVs) near nature protection areas, historical sites or other man-made structures is a difficult task. Traditionally, different sensing techniques are exploited with sonar being the most often used to extract depth information and to avoid obstacles. However, little has been published on complete control systems that utilize robotic vision for such underwater applications. This paper provides a proof of concept regarding a series of experiments investigating the use of stereo vision for underwater obstacle avoidance and position estimation. The test platform has been a ROV equipped with two industrial cameras and external light sources. Methods for underwater calibration, disparity map and 3D point cloud processing have been used, to obtain more reliable information about obstacles in front of the ROV. Results from laboratory research work and from field experiments demonstrate that underwater obstacle avoidance with stereo cameras is possible and can increase the autonomous capabilities of ROVs by providing appropriate information for navigation, path planning, safer missions and environment awareness.