Murat D. Aykin

On Feature Matching and Image Registration for Two‐dimensional Forward‐scan Sonar Imaging

The computer processing of forward-look sonar video imagery enables significant capabilities in a wide variety of underwater operations within turbid environments. Accurate automated registration of sonar video images to complement measurements from traditional positioning devices can be instrumental in the detection, localization, and tracking of distinct scene targets, building feature maps, change detection, as well as improving precision in the positioning of unmanned submarines. This work offers a novel solution for the registration of two-dimensional (2-D) forward-look sonar images recorded from a mobile platform, by optimization over the sonar 3-D motion parameters. It incorporates the detection of key features and landmarks, and effectively represents them with Gaussian maps. Improved performance is demonstrated with respect to the state-of-the-art approach utilizing 2-D similarity transformation, based on experiments with real data.

Dynamic scene analysis and mosaicing of benthic habitats by FS sonar imaging - Issues and complexities

Photo-mosaics generated automatically from as many as thousands of optical images have proved to be an effective technology to study the ecological patterns and dynamics of underwater ecosystems and benthic environments over spatial scales much larger than a single object or image. Unfortunately, optical systems, while useful in clear waters, are ineffective within environments with sources of turbidity and pollution, including lakes, marine sanctuaries, many ports and harbors. Two-dimensional high-resolution forward-scan imaging systems can serve as a suitable technology for constructing similar visual maps, provided that a range of complex imaging issues can be overcome. This paper investigates some of the complexities in analyzing dynamic events captured by a FS sonar imaging system when used in standard configuration to map the seafloor. Of special interest is the case of imaging targets at shorter ranges to maximize benthic object details. We give mathematical models that describe the dynamics associated with objects and shadows they cast on the seabed, and demonstrate some of these issues through examples from real data obtained in the lake on the University of Miami campus.

On feature extraction and region matching for forward scan sonar imaging

Automated processing of sonar video imagery enables valuable capabilities for a wide variety of underwater applications in turbid environments. Some key examples comprise the detection, localization and tracking of distinct scene targets, building feature maps, as well as improving positioning accuracy of unmanned submarines by means of image registration and 3D motion estimation to augment traditional positioning devices. This work offers a novel technique for the registration of 2D forward-look sonar images, by optimization over the sonar 3D motion parameters. It incorporates landmark detection through an adaptive clustering scheme with Gaussian map to represent key features at each frame. Improved performance is demonstrated in experiments with real data, both in terms of computation time and accuracy, relative to the state of the art, where the registration utilizes a simplified 2D image transformation model. Among many potentials, the method can improve precision in AUV navigation and environmental modeling.

Calibration and 3D reconstruction of underwater objects with non-single-view projection model by structured light stereo imaging.

Establishing the projection model of imaging systems is critical in 3D reconstruction of object shapes from multiple 2D views. When deployed underwater, these are enclosed in waterproof housings with transparent glass ports that generate nonlinear refractions of optical rays at interfaces, leading to invalidation of the commonly assumed single-viewpoint (SVP) model. In this paper, we propose a non-SVP ray tracing model for the calibration of a projector-camera system, employed for 3D reconstruction based on the structured light paradigm. The projector utilizes dot patterns, having established that the contrast loss is less severe than for traditional stripe patterns in highly turbid waters. Experimental results are presented to assess the achieved calibrating accuracy.

Modeling 2-D Lens-Based Forward-Scan Sonar Imagery for Targets With Diffuse Reflectance

Sonar images are formed by transmitting acoustical pulses and measuring the reflected sound power from the scene surfaces. The recorded signal by a sonar device encodes information about the shape and material properties of these surfaces. In this paper, we present a detailed derivation of an image model for a new class of high-resolution lens-based 2-D forward-looking sonar systems, when the diffuse reflectance of scene/target can be characterized by the Lambertian model. A simplified single patch model is generalized to account for the finite pulse width of the transmitted beam and the simultaneous arrival of scattered signals from multiple patches at the same range in a given azimuthal direction. Validating the model using intensity measurements of isolated cylindrical targets, we then demonstrate application for multipath reflections from bottom surfaces with cylindrical and spherical targets.

Three-Dimensional Target Reconstruction From Multiple 2-D Forward-Scan Sonar Views by Space Carving

Building 3-D object models from 2-D images is a key capability for target classification and identification, reacquisition, and environmental mapping, among many applications in underwater with poor visibility. We present a novel approach in utilizing multiple 2-D forward-look sonar images from known sonar poses to localize an acoustically opaque target and reconstruct its 3-D shape. Based on projections onto various images, the 3-D space not occupied by an imaged target within the sonar field of view is sequentially carved out, leaving the remaining volume as the estimate of the 3-D object region. The estimation generally improves with information from new distinct views, and moreover with images acquired through sonar roll motions, rather than circumnavigating the target. Computer simulations allow assessing the convergence properties and performance of the approach for convex and concave polygons. Additionally, results from experiments with real images of amorphous coral rocks and a miniature wood table demonstrate performance in the 3-D modeling of small objects with varying reflectance properties.

Efficient ray-casting of quadric surfaces for forward-scan sonars

A method is presented to effectively scan the 3-D surface(s) of a target comprising of quadric components for simulating the 2-D image acquired by a forward-scan sonar for a given pose. This involves a closed-form solution to compute the coordinates of a point on the target surface(s), projecting onto a given image pixel. A certain advantage over previous algorithms is we generate a minimal set of 3-D surface points for a given sonar viewpoint; therefore, it allows space and time efficiency for image synthesis.

Positioning and scene analysis from visual motion cues within water column

Over the last dozen or more years, many applications of vision-based positioning and navigation near the sea bottom and surface have been explored. Mid-water operations have primarily relied on traditional positioning systems, namely INS, DVL, gyros, etc. This paper investigates the application of a vision system for mid-water operations by exploiting stationary features within the water column. The unique nature of these environment - namely, the abundance of randomly distributed targets over a wide field of view and range of depth - are ideal for the application of well-known motion vision methods for 3-D motion estimation and scene analysis. We demonstrate through experiments with water tank and ocean data how various visual motion cues may be used for passive navigation, environmental assessment and target/habitat classification based on visual motion behavior.