Jeonghwe Gu

Development of passive acoustic landmark using imaging sonar for AUV's localization

We proposed and tested a passive acoustic landmark that can provide accurate navigation information for an autonomous underwater vehicle (AUV). The AUV used imaging sonar mounted on the vehicle to recognize shadows of the columntype landmark, and can measure vehicles own pose and relative position from the landmark which was set in a known position. The results of the proposed method had acceptable error (<;1 %) in the calculated position of the AUV. The proposed localization method will be useful to correct drift errors of inertial navigation systems for the AUV or to find the same position repeatedly.

Development of energy-harvesting unit and propulsion unit for a robotic buoy system

In this paper, we described about development of a robotic buoy for Internet of maritime thins (IoMT) technology. For IoMT, stand-alone data conllector is a key technology. The energy harvesting robotic buoy is proposed in the context. The proposed system is composed of three components: propulsion system, energy harvesting system, and buoyancy control system. The proposed system can hold its position within an uniform bound without mooring, and harvests wave energy. As a preliminary study, we developed hydrodynamic model for the whole system, and conducted simulation. Through the simulation, we implemented station keeping algorithm, and investigated optimal design parameters. To verify the simulation results we constructed prototype and carried out a series of experiments: station-keeping test, and power generation test. As a result, the whole system was proved that the propulsion system has sufficient manoeuvrability for station keeping within 3 m, and peak power generation was 8 W.

Robust Sonar-Based Underwater Object Recognition Against Angle-of-View Variation

Underwater object recognition based on deployed mobile nodes (underwater vehicles) is difficult, because the shape of an object in a sonar image is significantly changed depending on the direction from which the object is approached. The approaching directions of deployed mobile nodes cannot be predicted in advance. To solve the problem, the conventional underwater recognition algorithms use realistic template images generated by sonar image simulators, and compare the template images to the actual sonar image. However, the realistic sonar image simulation is computationally complex, and the comparison requires some preprocessing methods, such as image segmentation. To solve the problem, this paper proposes a sonar image simulator-based underwater object recognition algorithm. The proposed algorithms that is motivated by the generation mechanism of a sonar image can directly compare the actual sonar image and the simulation image generated by a simple sonar image simulator.

Experimental results of rapid underwater object search based on forward-looking imaging sonar

In case of marine accident, the rapid search for missing object such as an underwater wreckage is one of the most important tasks. For this task, an autonomous underwater vehicle (AUV) that equips forward-looking imaging sonar can be utilized. The search task is accomplished by the following strategy. In the first phase, the AUV scans the target underwater area in a fast speed, and detects some suspected objects. In the second phase, the AUV approaches to the object and investigates it to determine whether the suspected object is the target object to be found. This paper provides experimental results of underwater object search in a fast speed for realizing the first phase of the strategy.

On preprocessing methods for feature-based photo-mosaic underwater

In this paper, we studied some pre-process methods that make the feature matching faster, accurate and proper to be used in the photo mosaic process. Two properties and two image processing technique were considered: channel, size, contrast enhancement, and Gaussian smoothing. We investigated how they affect the feature quantity and matching result by controling related parameters. The smallest number of features were found in the red channel. As the image size decrease, the quantity of features and the computing time were decreased in the same manner in terms of ratio. Contrast enhancement methods made image features more distinct, but its amplification in the feature quantity was not that significant. The Gaussian smoothing increased the quantity of features in the way it has the maxima with an optimal sigma, blurring parameter. When the pre-processes are used in combination, the feature matching between pair images was improved 1.6 times more in quantity.

Sonar simulator-based underwater object recognition and its tolerance evaluation

The sonar images can be applied for underwater visual inspection using autonomous underwater vehicle (AUV) that requires underwater object recognition. However, the high noise and low resolution of sonar images are obstacles of object recognition. In this paper, we review a sonar image simulator-based underwater object recognition algorithm for forward-looking imaging sonar. Then, we verify the feasibility of the algorithm for AUV applications through evaluating its tolerance against AUV position error.

Real-time noise reduction for sonar video image using recursive filtering

The forward-looking imaging sonar is a prospective solution for underwater visual surveying because it allows longrange visibility in turbid water, and provides a high frame rate. However, the acoustic images are degraded by speckle noise. In this paper, we propose an algorithm to reduce the noise in a series of acoustic image frames obtained by using a forward-looking imaging sonar. The time-series model of the acoustic images are developed for predicting the changes in pixel coordinates. Also, the Kalman filter estimates the noise-reduced pixels of the images based on the acoustic image model. This recursive treatment is suitable for the successive image frames.

A method for automatic detection of underwater objects using forward-looking imaging sonar

A method for automatic detection or recognition using high-frequency forward-looking imaging sonar is proposed. The method includes segmentation of an underwater objects echo shape or its acoustic shadow shape. For accuracy of segmentation, some constraints were assumed on the underwater ground and the sonars pose. The separated shape was compared to simulated reference shapes to know its orientation or identity by the shape matching method that use shape context concept. An experiment conducted in a wave tank validated the method qualitatively. The method can be applied to an AUVs object-searching applications.

Sonar image processing algorithms for object recognition using hovering-type AUV “Cyclops”

In this paper, we describe the developed hovering-type AUV called “Cyclops” and discuss characteristics of imaging sonar DIDSON as an tool for AUV application. The Cyclops was designed to perform an advanced mission like object recognition, and its symmetric design enable to maximize the mobility of the vehicle. This hardware structure makes the maintenance of the vehicle fast and convenient. We introduce sonar image process algorithms from simple to advanced ones for AUV application. The algorithm includes segmentation for the extraction of reverberation shapes in sonar images, speckle reduction after segmentation, edge detection, and shape matching analysis.

Feasibility study of noise reduction for imaging sonar based on time-domain filtering

Underwater sonar image can be obtained by the multi-beam imaging sonar. Because of its high frame rate, the imaging sonar can be used for underwater survey especially in turbid water condition. However, the sonar image is generally degraded by the speckle noise. In this paper, we propose a realtime noise reduction algorithm by using the recursive least square algorithm, which is one of the adaptive filters. Through experiments, the performance of the proposed algorithm is analyzed and its feasibility is determined. Also the difference between the proposed algorithm and the conventional noise reduction algorithms that are based on spatial-domain filters are discussed.